How the adoption of drought-tolerant rice varieties impacts households in a non-drought year: Evidence from Nepal.
Table of contents Acknowledgements 4 About the authors 4 Abstract 7 1. Introduction 8 2. Data 9 3. Descriptive statistics 11 4. Estimating treatment effects 14 5. Results 19 5a. Impact of the adoption of STRVs on first-order outcomes 19 5b. Comparing results between long- and short-survey responses 23 5c. Impact of the adoption of STRVs on second-order outcomes 25 6. Conclusions 29 7. References 31 8. Appendix 32
1. Introduction
Farmers in unfavourable rice environments are vulnerable to weather shocks, especially as climate change makes weather patterns more extreme and less predictable. A common shock is drought, which can reduce crop yields and even result in total crop losses in rain-fed rice environments. Stress-tolerant rice varieties (STRVs) have been developed to reduce the impacts of climate shocks such as drought. STRVs are high yielding compared to traditional landrace varieties and reduce yield loss due to climate shocks compared to landrace varieties and other improved varieties. They thus have the potential to improve incomes and resilience to climate change. In addition, drought-tolerant varieties are also short-duration varieties. This can allow households to plant legumes or vegetables on their rice plots after the rice harvest, potentially improving nutrition and/or incomes, as households can consume or sell these crops. Between 2005 and 2008, several STRVs were validated and released in the drought-prone western development region of Nepal through a collaboration between the Nepal Agriculture Research Council (NARC), the Institute of Agriculture and Animal Science (IAAS), a local agricultural college and the International Fund for Agricultural Development (IFAD). Most of these varieties were bred to be tolerant to drought.
This paper estimates the impacts of the adoption of STRVs on first-order (mean yield, yield variation and growing duration) and higher-order impacts (labour, fertilizer use and the planting of vegetables and legumes) using data collected from 900 households in 2018, a non-drought year. The methods control for potential selection bias by estimating correlated random-effects (CRE) models that eliminate unobserved plot and household-level heterogeneity. In addition, this study examines how survey design affects household responses and estimated treatment effects by running an experiment in which half of the households receive a more detailed module on agricultural inputs.
Results find that even in a non-drought year, STRVs have a higher yield, reduced yield variance and shorter growing duration than local landrace varieties. Planting any kind of improved variety or hybrid induces households to apply more chemical fertilizer to plots, while STRVs and older improved varieties are planted to plots with more early-season chemical fertilizer and land preparation labour. Including more detailed input data collected from the longer survey version does not impact estimates of first-order treatment effects. It does allow for a more thorough examination of higher-order impacts of adoption, providing information on early-season fertilizer use and land preparation labour.
Our results have important implications for policymakers, and this study contributes to the literature in several ways. First, the first-order impacts of drought-tolerant varieties have not been widely studied, and the evidence is currently inconclusive. One study in India found that a popular drought-tolerant variety provides a yield advantage over other varieties, while another found that it reduces yields, even in a drought year (Yamano et al., 2018; Dar et al., 2020). Our results provide evidence that in a non-drought year in Nepal, STRVs offer a unique set of first-order benefits to farmers, increasing mean yield while reducing variance and growing duration. These benefits can increase incomes and reduce poverty via increased sales or reduced purchases of rice. Because rice is a staple crop for so many households, this could also improve food security. Combined with reduced yield losses in drought years, STRVs can substantially increase farmers' resilience to climate change. It is important for policymakers to understand the first-order impacts of the adoption of STRVs in typical non-drought years to evaluate returns on investments in STRVs. In addition, uncertainty about whether drought will occur may be a barrier to the adoption of STRVs. If farmers are aware that STRVs perform at least as well as other modern varieties during non-drought years, this could facilitate adoption.
Our findings on higher-order outcomes also greatly improve understanding of how the development of STRVs can affect rural communities. For instance, studies such as Mottaleb et al. (2017) that make ex-ante predictions of productivity increases due to STRVs can use our results to improve their predictions. According to Emerick et al. (2016), who find that adoption of a flood-tolerant variety in India causes higher-order effects such as increased fertilizer and labour use, these effects can arise for a few reasons. First is an income effect: STRVs provide higher expected yields on average, which increase households' expected incomes. This, in turn, may increase input use because farmers are wealthier. Second, there could be a marginal productivity effect if yields of STRVs are more responsive to inputs relative to other varieties. Finally, there could be a risk effect: farmers may be more willing to invest resources in a variety that is less likely to fail. The authors argue that it is crucial for policymakers to take these higher-order impacts of the adoption of STRVs into consideration when evaluating returns on investments. Thus far, there has been little research on higher-order impacts of drought-tolerant varieties, and none in Nepal. It is necessary to explore whether the impacts found by Emerick et al. hold in this context, particularly because drought-tolerant varieties have the added advantage of being short-duration varieties that could induce higher-order impacts such as growing legumes after the rice harvest (Yamano et al., 2018; Dar et al., 2020).
Our paper provides a methodological contribution to the literature by using CRE models to estimate treatment effects. These models eliminate household-level unobserved heterogeneity and, for first-order outcomes, plot-level unobserved heterogeneity. They subtract within-group unobserved heterogeneity in the same way as fixed-effects (FE) models, but unlike FE models they allow for the estimation of coefficients for variables with no within-group variation. This is an important contribution, as randomized data are not always available, and it improves on other commonly used methods that use observational data to estimate treatment effects. We argue that this approach leads to unbiased estimates more reliably than commonly used instrumental-variable approaches, as it is often difficult to find a valid instrument, and when it is possible, these are usually at the household level rather than the plot level. When households cultivate multiple fields, they are likely to target STRVs toward certain fields (for instance, drought-tolerant varieties are likely to be planted on plots prone to drought). Not accounting for this source of endogeneity could lead to selection bias and an inability to identify treatment effects. Panel data methods, as used in Yorobe et al. (2016), can eliminate this bias, but they require more than one year of data collection, which can be cost-prohibitive.
Finally, we contribute to the literature on survey design by examining the practical importance of collecting more detailed data on agricultural inputs when estimating treatment effects of an STRV. Previous studies provide evidence that data on labour are sensitive to survey design and recall period, while data on other inputs are not (Beegle et al., 2012; Dillon et al., 2012; Bardasi et al., 2011; Deininger et al., 2011; Arthi et al., 2018). Like these papers, we use an experimental method to examine the effect of survey design on collecting data on agricultural inputs. Our main contribution is testing how these different data affect results in two ways: first, the precision of first-order impacts when inputs are used as explanatory variables, and, second, whether having more granular data to use as dependent variables provides greater nuance regarding the higher-order impacts of the adoption of STRVs. This will help researchers determine whether they should devote data collection resources to conduct longer survey questionnaires.
The next section of the paper describes the data collected for this study. This is followed by a section explaining the empirical strategy, including detailed descriptions of outcome and explanatory variables. Next, we provide an overview of our results, and finally we offer conclusions to our work.
2. Data
This study uses data from household and community surveys conducted in 75 villages in Lamjung, Tanahu and Gorkha districts in November and December 2018. The 12 villages with a Consortium for Unfavorable Rice Environments (CURE) seed producer group (SPG) were selected, 12 villages that were adjacent to these villages were randomly selected, and 51 additional villages were randomly selected. These 51 villages were selected from an area that includes the Village Development Committees (1) (VDCs) in which the SPG villages are located, and all surrounding VDCs. The study area is represented by the yellow area in Figure 1. We sampled villages in this way so that we could estimate the impact of SPGs on the adoption of STRVs (findings from this study, as well as more detail regarding data collection, can be found in Vaiknoras et al., 2020) and to increase the likelihood of sampling a sufficient number of households that adopted STRVs.
Respondents from 12 randomly selected households were interviewed in each of the 75 selected villages for a total of 900 households. The survey included questions on household socio-economic characteristics, and GPS coordinates were collected for all households. The bulk of the survey collected detailed information on rice cultivation in the 2018 monsoon season (Nepal's main rice-growing season that runs from June to November) using three modules. The first of these identified each rice variety grown by the household in that season. Each variety was classified as either a traditional landrace variety, an old (released prior to 1990) modern improved variety (MV), a new MV (released in 1990 or later), an STRV, which is a subset of MVs, or a hybrid variety. This classification is important, as each of these variety types has a different yield potential.
The next module collected information on each plot cultivated by the household in that season, including the size of the plot and plot characteristics such as slope, whether it was irrigated during the 2018 monsoon season, and how prone to drought it is. Plots were defined as continuous parcels of land for which soil conditions were similar and input use was constant. For plots that had rice, additional questions on inputs were asked. Questions on labour asked about the application of labour per plot: how many people (household, paid and other unpaid) worked on the plot, the average number of days worked per person, and the average number of hours worked per day. Respondents were asked to consider all tasks for rice production on the plot from nursery and land preparation labour to threshing. This module also collected data on the quantity of organic fertilizer, chemical fertilizer and pesticides applied to all plots. All households answered these questions.
Half of the households, randomly selected, were administered a longer version of the questionnaire with an additional module that contained more detailed questions on labour and input use. This module asked the number of person-days by paid and unpaid men, paid and unpaid women and unpaid children spent on the following tasks (per plot): i) nursery, land preparation and planting; ii) weeding and pest control; iii) harvesting; and iv) threshing. It also asked more detailed questions on the application of fertilizer and pesticides: the number of times fertilizer was applied, the day in the season this occurred (with day 0 being the day of transplanting the rice seedlings into the field), the quantity per application, and for chemical fertilizer and pesticides, the type (diammonium phosphate (DAP), potash or urea for chemical fertilizer; herbicide, fungicide or insecticide).
Finally, the survey included a third module that linked the cultivated varieties to the households' plots and collected information at the variety-plot level. This included the quantity of seed of each variety planted on the plot, the area of the plot under which each variety was grown, and the quantity of grain of each variety harvested on the plot. This information allowed us to calculate yield, the multiplication ratio (quantity of seed harvested/quantity of seed planted) and seeding rate at the variety-plot level. For each variety on each plot, farmers also reported the age of seedlings on the day of transplantation, and the week and month of planting and harvesting, which allowed us to calculate growing duration at the variety-plot level. If households cultivated multiple varieties on one plot or cultivated the same variety on multiple plots, we identify these as separate observations. Plots were identified as contiguous parcels of land on which the same level of inputs was applied; therefore, all varieties in our data that were identified as being on a certain plot should have received the same amount of fertilizer, labour and pesticides.
The community survey interviewed village leaders regarding village services and amenities. Its purpose was to supplement the information collected at the household level. In particular, the community survey included questions about the presence of farmers' associations in the village and distance to extension services.
3. Descriptive statistics
To check whether survey length was truly assigned randomly to households, we compare descriptive statistics between households that received the long and short versions of the survey (Table A1). We look at general household descriptive statistics and characteristics that may have pushed enumerators to administer the shorter survey, such as those influencing difficulty reaching the household and survey length (e.g. number of cultivated plots). We find no differences in responses between assignment groups, indicating that assignment was random. Short-survey input responses also do not vary depending on survey assignment, providing additional evidence that assignment was random (Table 1). (2)
We compare short-survey responses to the sum of long-survey responses for fertilizer and labour to test whether they differ depending on how this information is collected (Table 1). The average quantity of organic and chemical fertilizer and pesticides does not vary by whether the household was administered a long or short version of the survey. Households applied about 8,600 kg/ha of organic fertilizer, 120 kg/ha of chemical fertilizer and 0.35 l/ha of pesticides according to both survey modules (pesticide use was low overall, and they were applied by only about 12 per cent of households). By contrast, estimates of labour use were far greater in the short-survey responses than long-survey responses. This could indicate that farmers are including labour for tasks beyond the ones specifically asked about or misunderstood the task categories. It could also mean that farmers are either over-reporting labour if asked about it generally or under-reporting if asked to remember labour by task, gender and unpaid vs. paid.
Our results are consistent with the literature. Beegle et al. (2012) found very little evidence of recall bias for fertilizer application but some bias for labour. They argue that fertilizer use may be less susceptible to recall bias than labour use because labour is used throughout the entire season, while fertilizer is generally applied only a few times. This holds true in our data; households that applied organic fertilizer or pesticides to their plots did so once on average per plot, and households that applied chemical fertilizer to their plots did so twice on average per plot. Other studies also found that the method of data collection and/or survey design (in particular, how labour tasks are defined) affected household responses for labour (Arthi et al. 2018; Dillon et al. 2012; Bardasi et al. 2011).
Households grow a mixture of rice variety types: 14 per cent of the household-plot-variety observations in our sample were local varieties; 19 per cent were old MVs; 24 per cent were non-STRV new MVs; 20 per cent were STRVs; and 23 per cent were hybrids (Table 2). STRVs were the most prevalent variety type (meaning there was more seed of STRVs planted to the plot than any other type) on 21 per cent of plots, compared to local varieties (9 per cent), old MVs (21 per cent), other new MVs (27 per cent) and hybrids (22 per cent).
The average multiplication ratio of STRVs is statistically identical to that of old and other new improved varieties, significantly higher than that of local varieties and lower than that of hybrid varieties. The seeding rate for STRVs was significantly higher than for hybrids and lower than for old MVs. The growing duration of STRVs is 17.72 weeks, which is nearly 2 weeks less than local varieties and also less than that of old and new MVs. All varieties were planted during the 11th week of the year, which corresponds to the first week of the month of Ashad, (3) but STRVs were planted later in the week on average than landraces, old and other new MVs. STRVs were harvested about a third of the way through the 29th week of the year, or the second week of Kartik, (4) about 2.5 days earlier than old MVs and over a week before local varieties. A majority of STRVs were planted on irrigated plots (73 per cent), although this is less than any other type of variety. A majority of STRVs (65 per cent) were grown on plots prone to drought, which is more than any other type of variety.
The average quantities of labour, organic and chemical fertilizer, and pesticides do not differ between plots on which STRVs are most prevalent and other plots (Table 3). Plots on which STRVs are most prevalent are not more likely to have been cultivated with legumes and/or vegetables once rice is harvested. Detailed input values from the long survey version also do not vary by STRV and other plots, except that households use more threshing labour on STRV plots (Table A2). This includes fertilizer use applied over the entire season and fertilizer applied on or before the day of transplantation, which is true for the majority of organic fertilizer and about half of chemical fertilizer.
In 2018, drought was not common but was more prevalent on plots that were primarily planted with STRVs than other plots: 8 per cent of STRV plots suffered from drought, compared to 3 per cent of other plots. In 2017, drought was more common but displayed the same pattern: 21 per cent of STRV plots suffered from drought, while 9 per cent of other plots suffered from drought. This is consistent with STRVs being more likely to be planted on plots that are prone to drought and not irrigated. There is variation in plot susceptibility to drought at the household level (not shown in the table): about 11 per cent of households have at least one plot that is prone to drought and at least one that is not prone to drought. About 54 per cent of households that reported that a plot suffered from drought in 2018 and 42 per cent of households that reported the same for 2017 also had at least one plot that did not suffer drought in the corresponding year.
4. Estimating treatment effects
We observe first-order outcomes [01.sub.ijk] (mean yield, yield variance and growing duration) of each variety k on plot j by household i:
[01.sub.ijk] = [[beta].sub.0] + [[beta].sub.1][T.sub.ijk] + [[beta].sub.2][S.sub.ijk] + [[beta].sub.3][P.sub.ij] + [[beta].sub.4][I.sub.ij.sup.l] + [[beta].sub.5][H.sub.i] + [[mu].sub.i] + [c.sub.ij] + [[epsilon].sub.ijk] (1)
Each combination of i, j, and k is unique, meaning that if household i grew two varieties on plot j, this is two observations. Similarly, if household i grew the same variety on two different plots, this is also two separate observations. If a household cultivated only one variety, then it would have only one observation. Most households (73 per cent) grew more than one rice variety in 2018, while 24 per cent of plots were cultivated under more than one variety. (5) The first-order outcome variables in equation 1 are mean yield, yield variance and growing duration. As a proxy for yield, we use the logged value of the multiplication ratio for variety k grown on plot j by household i. The multiplication ratio is the quantity of grain harvested in kg divided by the quantity of seed planted in kg. The benefit of using the multiplication ratio over yield is that the multiplication ratio does not require an accurate estimate of land size and may, therefore, be a more accurate measure of productivity. Farmer-reported plot sizes can be inaccurate, especially if the plots are on sloped land and non-standard units of measurement are used in reporting size, both of which are common in Nepal (Keita et al., 2010; Carletto et al. 2015). We use the logged value because the distribution of multiplication ratios is highly skewed to the right. We also examine equation 1 using quantity in kg/land planted in ha as the dependent variable to determine whether there are differences in results between the two measures of yield.
We also estimate the impact of the adoption of STRVs on yield variability over space. Following the moment-based approach developed by Antle (1983) and the estimation procedures described by Wossen et al. (2017), variability is measured as the squared estimation errors from equation 1 or 3. We compute the estimation errors after estimating the yield regression by subtracting the predicted outcome value by actual outcome value. Next, we square these errors and use them as the dependent variable, regressed on the same treatment and explanatory variables as the mean yield equation. The final first-order outcome is the duration of the growing season, measured as the number of weeks between when the rice seedlings were transplanted into the field and when rice was harvested.
[T.sub.ijk] refers to the seed type of variety k growing on plot j by household i; it includes four dummy variables, each of which equals 0 when the variety is a landrace type. The first dummy equals 1 when the variety is an old MV (released in 1990 or before); the second equals 1 when the variety is a new MV but not an STRV; the third equals 1 if the variety is an STRV; and the fourth equals 1 if the variety is a hybrid. We consider each variety type because each has a different potential yield, yield variance and growing duration. Under normal rainfall conditions, we expect that hybrids will have the highest yields, followed by STRVs and other new MVs, followed by older MVs and, finally, landraces. Since most farmers in our study area did not experience drought in the 2018 monsoon season, we do not expect the yield of STRVs to differ significantly from the yield of other MVs. We expect that the yield variance of STRVs will be lower than for traditional landraces, as their yields are less likely to vary due to variations in the availability of water, potentially even in a non-drought year. We do not expect other improved variety types or hybrids to have reduced variance. Because STRVs are bred to be short-duration varieties, they are expected to have a shorter growing duration than local varieties. Some other improved or hybrid varieties may have the same short-duration trait as STRVs, so they may also have a shorter growing duration.
[S.sub.ijk] in equation 1 refers to variables related to the seed and seedlings of variety k grown on plot j by household i. This includes the seeding rate (the rate of seeds planted on the plot in kg/ha) and age of seedlings in days on the day of transplantation, and a dummy variable equal to 1 if the seed for this variety was certified (and 0 if it came from recycled planting material of household i or another household). SPG member farmers in the area were trained to lower their seeding rates to increase yields, so we expect that higher seeding rates will reduce yields. Finally, it also includes the area on plot j in ha on which variety k was cultivated.
Variables in [P.sub.ij] refer to the plot characteristics of plot j. The slope variable is a dummy variable equal to 0 if the plot is flat and 1 if it has a gentle, moderate or steep slope. We also include a dummy variable equal to 1 if the respondent reported that the plot is susceptible to drought, and 0 otherwise. The irrigation dummy variable is equal to 1 if the plot is irrigated, and 0 otherwise.
[I.sub.ij.sup.l] is a vector of inputs applied on plot j. When the superscript l = 1, the vector includes input variables computed from the short version of the questionnaire, and when l = 2, the vector includes inputs measured from the long survey version. Both vectors include a dummy variable equal to 1 if the household applied pesticides (6) to plot k. Vector l = 1 also includes the total quantity of organic fertilizer and chemical fertilizer applied per ha to plot k, and the total amount of labour in person-days/ha applied to plot k. Vector l = 2 instead includes the quantity of organic fertilizer (kg/ha), which is the sum of all applications collected from the longer survey, the quantity of urea, DAP and potash (kg/ha) each, and four variables to capture labour: land preparation, weeding, harvesting and threshing, each in person-days/ha on the plot. These are each aggregated over paid and unpaid and over men, women and children. (7) We expect that higher quantities of fertilizer and labour will increase yields.
Household-level variables ([H.sub.i]) included in the models are the sex of the household head (1 = female) and a dummy variable equal to 1 if the household head is literate. We also include the elevation of the household dwelling in metres above sea level (masl). Finally, this vector includes a dummy variable equal to 1 if the household resides in a village where there is an SPG, which has been found to increase adoption of STRVs and use of some best management practices that could affect the outcomes of interest (Vaiknoras et al., 2020).
Equation 1 is a random-effects (RE) model that assumes that additional unmeasured factors exist that affect outcomes of interest. Some of these factors, such as unobserved farmer ability, vary only across households and are included in [[mu].sub.i]. Others vary within households and across plots, such as soil quality; these are included in [c.sub.ij]. All remaining factors that might also vary across variety-plot observations of the seed of variety k grown on plot j (such as quality of seed) are included in the error term [[epsilon].sub.ijk]. Unobserved household and plot characteristics are likely to be correlated with the adoption of STRVs, biasing treatment effect estimates. We eliminate this bias by de-meaning our data to the greatest extent possible and estimating CRE models which produce within-group effects that are identical to FE models but also estimate between-effects (Schunck, 2013):
[01.sub.ijk] = [[beta].sub.0] + [[beta].sub.1][T.sub.ijk] + [[beta].sub.2][S.sub.ijk] + [[beta].sub.3][P.sub.ij] + [[beta].sub.4][I.sub.ij.sup.l] + [[beta].sub.5][H.sub.i] + [[pi].sub.1][[bar.T].sub.lj] + [[pi].sub.2][[bar.S].sub.lj] + [[pi].sub.3][[bar.P].sub.l] + [[pi].sub.4][[bar.I].sub.l] + [v.sub.ij] +[m.sub.i] + [[epsilon].sub.ijk] (2)
Equation 2 includes the plot-level means [[bar.T].sub.ij] and [[bar.S].sub.ij] of the variables in [T.sub.ijk] and [S.sub.ijk]. This removes the between-plot effects of these variables, including those coming from unobserved plot characteristics such as soil quality. Thus, [[beta].sub.1] and [[beta].sub.2] are estimates of the within-plot effects of variables [T.sub.ijk] and [S.sub.ijk], respectively, free of plot-level selection bias that arises when households target varieties or practices towards plots with certain characteristics. Coefficients [[pi].sub.1] and [[pi].sub.2] measure the between-plot effects of variables in [T.sub.ijk] and [S.sub.ijk], respectively. Adding [[bar.P].sub.i] and [[bar.I].sub.l], which represent household means for plot-level characteristics and inputs used, respectively, removes the between-household effects for these variables, rendering [[beta].sub.2] and [[beta].sub.3] estimates of their within-household effects. This ensures that the estimated coefficients in [[beta].sub.2] and [[beta].sub.3] are free from household selection bias. Coefficients [[pi].sub.3] and [[pi].sub.4] measure the between-household effects of variables [P.sub.ij] and [I.sub.ij.sup.l]. [[beta].sub.5] remains the between-household effect of household-level variables in [H.sub.i].
Second-order outcomes [02.sub.ij] (input use and legume/vegetable cultivation) are observed for each plot j cultivated by household i. The RE version of these models is:
[02.sub.ij] =[[beta].sub.0] + [[beta].sub.1][T.sub.ij] + [[beta].sub.3][P.sub.ij] + [[beta].sub.5][H.sub.i] + [[mu].sub.i] + [c.sub.ij] (3)
Second-order outcomes in [02.sub.ij] include total quantity of organic fertilizer in kg/ha, total quantity of chemical fertilizer in kg/ha and total person-days of labour applied per hectare on plot j. We do not estimate the effect on pesticides because pesticide use was so low in our sample. We hypothesize that farmers who grow an STRV will use more fertilizer and more labour due in part to the reduction in the risk of yield loss due to drought. For inputs applied at the start of the season, this should hold true regardless of whether there is drought that season, because farmers do not yet know what weather conditions will be, and we expect that households are more willing to invest in inputs for crops that are less likely to fail. As the season continues, farmers observe the weather conditions. In the case of drought, we expect STRVs to receive far more inputs than non-STRVs, as the likelihood of non-STRV crop failure increases. In the case of normal weather/no drought, input use for STRVs and non-STRVs should equalize as the season continues, as risk of crop failure due to drought lowers and converges to zero for all varieties. While 2018 was not a drought year, the vast majority of organic fertilizer and about half of chemical fertilizer were applied at the start of the season, so we expect to see effects of the adoption of STRVs on those outcomes. To investigate further, we estimate effects on early-season fertilizer applications and land preparation labour using responses from the long survey version.
Because STRVs are short-duration varieties, their adoption provides households with an opportunity to cultivate legumes or vegetables on the plot once rice is harvested (legumes and vegetables were combined because fewer than 2 per cent of plots had vegetables cultivated on them). Therefore, we also estimate the impact of adoption of STRVs on the probability that a household grows legumes and/or vegetables on the plot during the monsoon season.
In equation 3, the treatment variable vector [T.sub.ij] again refers to variety type as in equations 1 and 2. Because plots may have had more than one variety cultivated on them, this represents the variety type that has the greatest quantity of seed planted on the plot, using a series of dummy variables for which 0 = landrace for each. For the first dummy variable, 1 equals old MV; for the second, 1 equals new MV but not STRV; for the third, 1 equals STRV; and for the fourth, 1 equals hybrid. [P.sub.ij] and [H.sub.i] are the same plot- and household-level variables included in equations 1 and 2, with the addition of plot size in [P.sub.ij].
To estimate the CRE model for equation 3, [[bar.T].sub.i] and [[bar.P].sub.i] are included, which eliminate between-household effects of variety choice and plot characteristics:
[02.sub.ijk] = [[beta].sub.0] + [[beta].sub.0][T.sub.ij] + [[beta].sub.3][P.sub.ij] + [[beta].sub.5][H.sub.i] + [[pi].sub.1][[bar.T].sub.l] + [[pi].sub.3][[bar.P].sub.l] + [m.sub.i] + [c.sub.ij] (4)
CRE models remove major sources of potential bias in the treatment effect estimates, but they do not eliminate all bias. In equation 2, there could remain factors that vary over i, j and k in [[epsilon].sub.ijk] that could bias results. For example, seed of certain varieties could be of higher quality than others. We include variety-level variables to control for as much of this heterogeneity as possible, reducing the likelihood of bias. In model 4, there is a greater chance of bias because unobserved plot-level heterogeneity remains. Even if households were to target certain varieties towards plots with specific unobserved characteristics, treatment effects in model 2 ([[beta].sub.1]) would not be biased; however, the treatment effects in model 4 ([[beta].sub.2]) could be biased. In addition, the CREs do not eliminate bias that could be present in [[beta].sub.5], [[pi].sub.1], [[pi].sub.2], or [[pi].sub.3], as they can still be correlated with the remaining error terms [v.sub.ij] and [m.sub.i]. This does not affect our treatment estimates in either model but means that household-level covaraites and plot and household between effects should not be interpreted as causal.
[[beta].sub.1] and [[beta].sub.2] from model 2 are estimated using only within-plot variation, and they lose efficiency when there is little within-plot variation. Likewise, [[beta].sub.3] and [[beta].sub.4] from models 2 and 4 only use within-household variation. We examine whether these households and plots are representative of all households and plots and find that they differ in several ways (Tables A3 and A4). Households that grew more than one variety in 2018 grew 2.37 varieties on average and have an older head of household on average and a greater number of household members. They live farther away from roads and at a higher elevation. Not surprisingly, they cultivated a greater number of plots in 2018. Plots with more than one variety were larger, less likely to be sloped, and have less labour and less fertilizer applied per ha. These differences mean that our results from models 2 and 4 are not necessarily representative of the greater population; this is important to keep in mind as we interpret our estimates.
We perform an augmented regression test on the statistical significance of [[pi].sub.1], [[pi].sub.2], and [[pi].sub.3] to test whether the between-group estimate is significantly different from the within-group estimate (Baltagi, 2008). If they are not significantly different (if [[pi].sub.1], [[pi].sub.2], or [[pi].sub.3] is not statistically significant), then selection bias does not affect our results and the RE model is valid. It is also likely to be more efficient than the CRE model.
5. Results
5a. Impact of the adoption of STRVs on first-order outcomes
To obtain the most precise yield estimates possible, we first estimate the first-order impact equations using the more basic input vector (l = 1) for all households (Table 5). Next, to compare the precision of first-order estimates between the basic (l = 1) and more detailed (l = 2) input vectors, mean yield, yield variance and growing duration equations are estimated with only the sub-sample of households that received the longer survey questionnaire.
The coefficients on the means of each variety type for all three first-order outcome variables are not significant, indicating that between effects do not vary significantly from within effects. Thus, our treatment effects are not driven by selection bias, and the RE models are valid, so we discuss our results as ranges between the two estimates. Estimated STRV treatment effects across RE and CRE results are similar, indicating that relying only on plots with more than one variety does not affect results by much. Compared to landraces, old MVs increase yield by 21 per cent, new (non-STRV) MVs increase yield by 28 per cent, STRVs by 27 per cent according to RE results and by 30 per cent according to CRE results, and hybrids by 124 per cent according to RE results and by 119 per cent according to CRE results. Each of these effects is statistically significant at 1 per cent for RE results; STRV and hybrid coefficients are also statistically significant at 1 per cent in the CRE model. These effects all represent the within-plot effects--i.e. the effect of growing an STRV or other variety type vs. a landrace on a particular plot. In the RE model, the 95 per cent confidence intervals for STRVs overlap those of old and new MVs, providing evidence that the yield gain of STRVs is equivalent to that of old and other new MVs.
STRVs reduce the squared residuals from the yield equation, which represents yield variance, by 0.20 to 0.45 from local landraces (representing a reduction of about 50-100 per cent); this is significant at 5 per cent for both RE and CRE models. We find no evidence that other MVs or hybrids reduce yield variance relative to landraces.
Table A1 shows results for which the dependent variable is mean yield and yield variance, where yield is measured as quantity of rice harvested in kg/area planted in ha. Results for impacts of STRV and other MVs on mean yield are similar to those in Table 5, while coefficients for hybrids are much smaller (though still significant). For yield variance, STRV results are similar, but now in the RE results, other new MVs also reduce variance.
STRVs reduce the number of weeks between seedling transplantation and harvest by 1.26-1.32 weeks (significant at 1 per cent). Similarly, hybrids reduce this time by 1.07-1.51 weeks (significant at 1 per cent). Non-STRV new MVs reduce this duration by 0.64-0.65 weeks (significant at 1 per cent). The 95 per cent confidence intervals for STRVs overlap those of other new MVs and hybrids. There is no evidence that old MVs have a shorter growing duration than landraces.
Other variety-specific characteristics are significant as well. Increasing the seeding rate reduces yields by 0.7-0.8 per cent for every additional kg/ha of seed planted. Using certified seed increases yield by 17 per cent; this is not significant in the CRE model, but the plot mean is also not significant, indicating that the RE result is valid. An additional kg/ha in the seeding rate increases yield variance by 0.001 according to RE results. Being planted on an additional hectare of land increases yield by 39 per cent. However, plots on which more land is cultivated reduce yield by 81 per cent for each additional hectare. This indicates that larger plots produce lower yields, but for an individual variety, being planted on more land increases yield. It could be that farmers plant larger areas with varieties they believe will be high yielding.
Several plot-level variables are significant determinants of yield in the RE but not CRE models: in the RE model, an additional kg/ha of chemical fertilizer increases yield by 0.1 per cent, growing rice on a sloped field reduces yield by 7.5 per cent, and growing rice on an irrigated field increases yield by 11.8 per cent. While the CRE results are not significant for these variables, neither are their household means, providing evidence that RE results are valid. An additional kg/ha of chemical fertilizer has a small effect on variance, reducing the squared residuals of mean yield by 0.001. An additional kg/ha of chemical fertilizer reduces growing duration by 0.002 weeks according to RE results. Being planted on an irrigated field increases the duration by 0.55 weeks according to RE results. While slope increases the duration in the RE model results, it is not significant in the CRE model, while the household mean slope is significant in the CRE model. Thus, we conclude that slope does not have a true effect on the duration.
Finally, households with a literate head of household achieve durations that are 0.39 weeks shorter than households with a non-literate head of household. An additional masl reduces yield by 0.02 per cent and increases growing duration by 0.001 weeks. Living in a village with an SPG increases yield by about 15.4 per cent.
5b. Comparing results between long- and short-survey responses
Table 6 presents mean yield, yield variance and growing duration RE results for the sub-sample of households that were randomly assigned to the longer version of the household questionnaire. Only RE results are presented because the plot means of the treatment variables in the CRE (presented in Table 5) are not significant, and reducing the sample by half already reduces the variance in the results. Columns 1-3 present results using the basic input vector (l = 1), while columns 4-6 present results using the more detailed input vector (l = 2) explained in Table 1. We include these two sets of results to provide a valid comparison, since standard errors are expected to increase due to the reduction in sample size alone.
Reducing the sample by half but leaving the covariates the same as the models in Table 5 changes some of the coefficients and standard errors of the treatment variable; in particular, old MVs no longer have an effect on yield but do have a negative effect on yield variance. In general, we consider the results from Table 5 more reliable because they use a larger sample. Comparing the results from columns 1-3 and columns 4-6, we see that adding the more detailed input variables has a negligible effect on our treatment effect estimates; coefficients change only slightly. Thus, obtaining the more detailed input variables makes no real difference in estimating the treatment effects of the adoption of STRVs on mean yield, yield variance and season duration. In addition, we gain little insight into the role of inputs on our first-order outcomes; in the models where l = 2, urea and DAP have similar effects on yield and yield variance as chemical fertilizer does when l = 1, indicating that we are able to assess the role of chemical fertilizer with only basic questions. Potash is not significant, which could be meaningful; estimating the role of nutrients individually could provide an indication of which nutrients are needed more than others. Only one of the labour variables is significant; threshing is negatively associated with growing duration. Early harvesting could leave households more time for threshing at the end of the season. (8)
5c. Impact of the adoption of STRVs on second-order outcomes
Plots cultivated mainly under STRVs (9) have between 21 and 44 additional kg/ha of chemical fertilizer applied to them compared to plots cultivated under landraces (Table 7). Households apply between 25 and 41 additional kg/ha of chemical fertilizer to plots cultivated under hybrids, new MVs and old MVs compared to those under landraces; the 95 per cent confidence interval of STRVs overlaps those of these other variety types. We find no effect of cultivation of STRVs (or hybrid or other MVs) on labour or organic fertilizer use. In contrast to previous research (Yamano et al., 2018; Dar et al., 2020), we find no effect on legume/vegetable cultivation (Table 7). Although we expected to observe effects of the adoption of STRVs on organic fertilizer use and legume/vegetable cultivation, the descriptive statistics shed light on why these treatment effects are not significant. Households apply a large amount of organic fertilizer to all of their plots regardless of variety type. In addition, although households do achieve shorter growing durations with STRVs compared to landraces (Table 5), this reduction is split between slightly later planting and slightly earlier harvesting (Table 2).
Other plot characteristics also correlate with second-order outcomes. While irrigation and being prone to drought have a positive affect on labour in the RE model, neither is significant in the CRE results, indicating that households with irrigated and drought-prone fields tend to use more labour. Sloped fields are more likely to be cultivated under legumes or vegetables according to RE results, but this is not significant in CRE results. The larger the plot, the less labour and organic and chemical fertilizer applied in both RE and CRE models.
Household characteristics also affect second-order outcomes (though we are not able to explore the potential endogeneity of these characteristics). As elevation rises, households apply less labour, less organic fertilizer and less chemical fertilizer to their plots. Households in SPG villages apply less labour and less chemical fertilizer but are more likely to cultivate legumes and vegetables, which is consistent with Vaiknoras et al. (2020). Finally, women farmers are less likely to cultivate legumes and vegetables, while literate farmers are more likely to do so.
Higher-order outcomes are investigated in more depth using the detailed input variables from the long survey to investigate early-season inputs more clearly (Table 8). Households apply more chemical fertilizer on or before the day of transplantation to STRV plots than local variety plots by 17-29 kg/ha, significant at 5 per cent in both the RE and CRE models. According to RE results, old MVs get 21 kg/ha more early-season chemical fertilizer than landraces. This is not significant in the CRE results. Households apply 22 additional days of labour to STRV plots, and 24 to old MV plots, both significant at 5 per cent in the RE model. In the CRE results, STRV plots receive 30 additional person-days of labour, but this is significant only at 10 per cent. These increases can be attributed to increases in male labour: according to RE and CRE results, STRVs receive 13-24 additional hours of male land preparation labour, and old MVs receive 14-17 (though the CRE results for old MVs are significant only at 10 per cent).
6. Conclusions
STRVs are bred to help rice farmers mitigate climate shocks by reducing yield variability that arises from such shocks. In the mid-hills region of Nepal, where drought is the most significant climate concern, the adoption of STRVs has become common. This study estimates that about 20 per cent of the rice seeds planted in the 2018 monsoon season in the study area were drought-tolerant varieties. STRVs increase yield, reduce yield variance and reduce the growing duration compared to local landraces, outperforming them even in a non-drought year. Furthermore, there is no yield penalty for STRVs relative to other MVs in a non-drought year. This information can help policymakers allocate resources to the development of different varieties. It is also encouraging for adoption; if farmers are aware of the benefits of adoption in non-drought years, they may be more likely to adopt. This would offer additional protection in drought years, improving their resilience to climate shocks.
The higher-order outcomes of the adoption of STRVs and other MVs and hybrids have implications for agriculture in unfavourable rice environments. Households apply more fertilizer to STRV plots and to other MV and hybrid plots. Because only STRVs are expected to significantly reduce the risk of cultivation, this increase in fertilizer use is likely due to income or marginal productivity effects, as explained in Emerick et al. (2016). This implies that increased adoption of any of these variety types will induce households to increase fertilizer use, which could further increase yields. Only STRVs and old MVs have increased early-season fertilizer use and land preparation labour. For STRVs, this could be due to risk reduction or short growing duration, but old MVs have neither of those traits, so it is difficult to identify why they also receive more of these inputs.
We found no evidence that the adoption of STRVs or other varieties increases a household's likelihood of growing legumes and/or vegetables. If policymakers wish to increase cultivation of legumes or vegetables, adopters of STRVs and hybrid varieties could be educated about planting them after harvesting short-duration rice varieties. This may be impactful, given that other indicators of knowledge (being literate and living in an SPG village) make households more likely to cultivate legumes and/or vegetables. The adoption of STRVs would then have additional impacts on nutrition and/or income, as households could consume or sell their legume and/or vegetable harvest.
This study demonstrates how CRE models can be used to eliminate plot- and household-level selection bias in areas where farmers commonly grow multiple varieties on a plot and/or grow the same variety across different plots. With enough plot-level variation, unobserved plot heterogeneity can be eliminated to control for plot selection bias. In areas with little plot-level variation, household CREs can still be used to control for household selection bias if households grow multiple varieties per season. This provides researchers with a valid way to estimate treatment effects that does not require randomized data or an instrumental variable.
The experiment to randomize survey design offers insights for future researchers. Despite the differences in reported labour data across short and long survey questionnaires, this did not affect first-order treatment effects when added as covariates. Therefore, researchers may not need to collect very detailed data on inputs if their main goal is to estimate first-order outcomes. The main benefit of collecting more granular data was that it allowed more nuanced exploration of higher-order outcomes of the adoption of STRVs; without these data we would not have known the impacts of the adoption of STRVs on early-season fertilizer use or land preparation labour.
These results are important for researchers and policymakers who evaluate the impacts and returns on investment of STRVs. Because farmers do not experience drought in most years, knowing the performance of STRVs in non-drought years is crucial. More research is needed to estimate the effects of STRVs in drought years; in particular, it is important to know how STRVs perform relative to hybrids in drought years, since hybrids have higher mean yields than STRVs in non-drought years. Our results, along with further research, will provide policymakers with evidence of the benefits of STRVs in Nepal and other countries. This is crucial for developing and promoting technologies to help farmers adapt to a changing climate.
7. References
Antle, J., 1983. Testing the stochastic structure of production: a flexible-moment based approach. Journal of Business and Economics Statistics 1, 192-201.
Arthi, V., Beegle, K., De Weerdt, J., Palacios-Lopez, A., 2018. Not your average job: Measuring farm labor in Tanzania. Journal of Development Economics 130, 160-172.
Baltagi, B.H., 2008. Econometric Analysis of Panel Data. Wiley, New York, NY.
Bardasi, E., Beegle, K., Dillon, A., Serneels, P., 2011. Do Labor Statistics Depend on How and to Whom the Questions are Asked? Results from a Survey Experiment in Tanzania. The World Bank Economic Review 25, 418-447.
Beegle, K., Carletto, C., Himelein, K., 2012. Reliability of recall in agricultural data. Journal of Development Economics 98, 34-41.
Carletto, C., Jolliffe, D., Banerjee, R., 2015. From Tragedy to Renaissance: Improving Agricultural Data for Better Policies. The Journal of Development Studies 51, 133-148.
Dar, M.H., Waza, S.A., Shukla, S., Zaidi, N.W., Nayak, S., Hassain, M., Kumar, A., Ismail, A.M., Singh, U.S., 2020. Drought Tolerant Rice for Ensuring Food Security in Eastern India. Sustainability 12.
Deininger, K., Carletto, C., Savastano, S., Muwonge, J., 2012. Can diaries help in improving agricultural production statistics? Evidence from Uganda. Journal of Development Economics 98, 42-50.
Dillon, A., Bardasi, E., Beegle, K., Serneels, P., 2012. Explaining variation in child labor statistics. Journal of Development Economics 98, 136-147.
Emerick, K., De Janvry, A., Sadoulet, E., Dar, M.H., 2016. Technological Innovations, Downside Risk, and the Modernization of Agriculture. American Economic Review 106, 1537-1561.
Keita, N., Carfagna, E., Mu'Ammar, G., 2010. Issues and guidelines for the emerging use of GPS and PDAs in agricultural statistics in developing countries., The Fifth International Conference on Agricultural Statistics (ICAS V), Kampala, Uganda.
Mottaleb, K.A., Rejesus, R.M., Murty, M.V.R., Mohanty, S., Li, T., 2017. Benefits of the development and dissemination of climate-smart rice: ex ante impact assessment of drought-tolerant rice in South Asia. Mitigation Adaptation Strategies Global Change 22, 879-901.
Schunck, R., 2013. Within and between estimates in random-effects models: Advantages and drawbacks of correlated random effects and hybrid models. The Stata Journal 13, 65-76.
Vaiknoras, K., Larochelle, C., Alwang, J., 2020. The spillover effects of seed producer groups on non-member farmers in mid-hill communities of Nepal IFAD Research Series. IFAD, Rome.
Yamano, T., Dar, M.H., Architesh, P., Ishika, G., Malabayabas, M.L., Kelly, E., 2018. The impact of adopting risk-reducing, drought-tolerant rice in India. Impact Evaluation Report. International Initiative for Impact Evaluation (3ie), New Delhi.
Yorobe Jr., J.M., Ali, J., Pede, V.O., Rejesus, R.M., Velarde, O.P., Wang, H., 2016. Yield and income effects of rice varieties with tolerance of multiple abiotic stresses: the case of green super rice (GSR) and flooding in the Philippines. Agricultural Economics 47, 261-271.
8. Appendix
Table A1. Means (standard errors) of household characteristics, short-
version vs. long-version survey responses
Variable Households assigned to
short survey only mean (std.
dev.)
Age of head of household 51.68 (13.12)
Sex of head of 0.23 (0.42)
household (1 = female)
Head of household 0.74 (0.44)
is literate (1 = yes)
Number of household 3.85 (1.47)
members
Wealth index -0.11 (1.55)
Distance to road (m) 43.44 (60.66)
Elevation (masl) 644.31 (212.71)
Number of plots 1.58 (0.67)
cultivated in 2018
Number of rice varieties 1.78 (0.94)
cultivated in 2018
Number of observations 492
Variable Households assigned to long
survey mean (std. dev.)
Age of head of household 52.38 (13.43)
Sex of head of 0.21 (0.41)
household (1 = female)
Head of household 0.69 (0.46)
is literate (1 = yes)
Number of household 3.89 (1.38)
members
Wealth index 0.04 (1.54)
Distance to road (m) 45.49 (66.69)
Elevation (masl) 650.58 (206.89)
Number of plots 1.62 (0.64)
cultivated in 2018
Number of rice varieties 1.73 (0.90)
cultivated in 2018
Number of observations 406
Note: There were no statistically significant differences between
groups. Two households were missing several responses so were not
included in the table. Estimates for age, sex, literacy of the head of
household were based on 880 total responses, while wealth index is
based on 891 total responses due to missing observations.
Table A2. Means (standard deviations) of detailed inputs on plots where
STRVs are the most prevalent kind vs. all other plots
Variable Plots on which STRVs
are the most prevalent
type
Organic fertilizer (kg/ha) 8,164.91 (9,282.59)
Organic fertilizer (kg/ha) applied 7,420.15 (9,180.39)
on or before day of
seedling transplantation
Number of times organic 1.01 (0.11)
fertilizer was applied (for
households that applied
organic fertilizer)
Urea (kg/ha) 69.27 (62.01)
DAP (kg/ha) 44.44 (46.68)
Potash (kg/ha) 3.65 (10.22)
Chemical fertilizer (sum 63.51 (69.07)
of urea, DAP and potash)
(kg/ha) applied on or
before day of seedling
transplantation
Number of times chemical 2.06 (0.66)
fertilizer was applied (for
households that applied
chemical fertilizer)
Total hired labour 138.77 (165.48)
(person-days/ha)
Total unpaid labour 208.45 (188.49)
(person-days/ha)
Total land preparation 130.64 (92.48)
labour (person-days/ha)
Total weeding labour 76.87 (59.90)
(person-days/ha)
Total harvesting 74.37 (53.52)
labour (person-days/ha)
Total threshing labour 64.54 (50.36)
(person-days/ha)
Number of observations 103
Variable Plots on which non-STRVs
are the most prevalent type
Organic fertilizer (kg/ha) 7,864.10 (10,853.56)
Organic fertilizer (kg/ha) applied 7,348.58 (10,155.34)
on or before day of
seedling transplantation
Number of times organic 1.03(0.16)
fertilizer was applied (for
households that applied
organic fertilizer)
Urea (kg/ha) 74.21 (62.01)
DAP (kg/ha) 44.32 (54.97)
Potash (kg/ha) 6.44 (12.11)
Chemical fertilizer (sum 63.51 (59.76)
of urea, DAP and potash)
(kg/ha) applied on or
before day of seedling
transplantation
Number of times chemical 2.19 (0.83)
fertilizer was applied (for
households that applied
chemical fertilizer)
Total hired labour 143.70 (405.12)
(person-days/ha)
Total unpaid labour 190.58 (169.15)
(person-days/ha)
Total land preparation 116.94 (82.44)
labour (person-days/ha)
Total weeding labour 79.55 (65.94)
(person-days/ha)
Total harvesting 66.03 (46.75)
labour (person-days/ha)
Total threshing labour 55.44 (38.47) (**)
(person-days/ha)
Number of observations 439
Note: (**) denotes statistical significance at 5 per cent.
Table A3. Mean (std. dev.) household characteristics of households that
grew one vs. more than one rice variety
Variable Mean (std. dev.) households
that grew one variety
Age of head of household 50.42 (12.83)
Sex of head of 0.23 (0.42)
household (1 = female)
Head of household 0.74 (0.44)
is literate (1 = yes)
Number of household 3.74 (1.39)
members
Wealth index 0.04 (1.59)
Distance to road (m) 38.57 (52.08)
Elevation (masl) 620.65 (198.15)
Number of plots 1.26 (0.45)
cultivated in 2018
Number of rice varieties 1.00 (0.00)
cultivated in 2018
Variable Mean (std. dev.) households
that grew more than one
variety
Age of head of household 53.24 (13.46) (***)
Sex of head of 0.22 (0.41)
household (1 = female)
Head of household 0.70 (0.46)
is literate (1 = yes)
Number of household 3.96 (1.45) (**)
members
Wealth index -0.11 (1.51)
Distance to road (m) 48.87 (70.90) (**)
Elevation (masl) 668.24 (216.87) (***)
Number of plots 1.87 (0.68) (***)
cultivated in 2018
Number of rice varieties 2.37 (0.84) (***)
cultivated in 2018
Note: (*/**/***) denotes statistical significance at 10, 5 and 1 per
cent, respectively.
Table A4. Mean (std. dev.) plot characteristics of plots with one vs.
more than one variety
Variable Mean (std. dev.) plots with one variety
Labour (person-days/ha) 633.09 (556.53)
Organic fertilizer (kg/ha) 9,199.65 (10,281.75)
Chemical fertilizer (kg/ha) 126.58 (116.29)
Grew legumes and/or 0.32 (0.47)
vegetables in
monsoon season (1 = yes)
Plot is sloped 0.74 (0.44)
Plot is irrigated 0.85 (0.36)
Plot is prone to drought 0.51 (0.50)
Size of plot (ha) 0.21 (0.16)
Variable Mean (std. dev.) plots with
more than one variety
Labour (person-days/ha) 498.58 (464.93) (***)
Organic fertilizer (kg/ha) 7,024.65 (8,424.01) (***)
Chemical fertilizer (kg/ha) 102.93 (83.86) (*)
Grew legumes and/or 0.33 (0.47)
vegetables in
monsoon season (1 = yes)
Plot is sloped 0.60 (0.49) (***)
Plot is irrigated 0.88 (0.32)
Plot is prone to drought 0.48 (0.50)
Size of plot (ha) 0.34 (0.28) (***)
Note: (*/**/***) denotes statistical significance at 10, 5 and 1 per
cent, respectively.
Table A5. Coefficient (standard error) of RE and CRE estimates of the
impact of variety type and covariates on mean yield and yield variance
using yield (quantity harvested in kg/area planted in ha) as dependent
variable
(1) (3)
Variables Yield mean: Plot Yield mean: Plot
RE CRE
[T.sub.ijk]
Variety type
Old MV 0.333 (***) 0.254 (**)
(0.064) (0.111)
New MV (not 0.392 (***) 0.308 (***)
STRV) (0.059) (0.093)
STRV 0.333 (***) 0.414 (***)
(0.060) (0.109)
Hybrid 0.614 (***) 0.685 (***)
(0.065) (0.106)
[[bar.T].sub.ij]
Mean old MV 0.108
(0.126)
Mean new MV 0.118
(not STRV)
(0.108)
Mean STRV -0.097
(0.126)
Mean hybrid -0.087
(0.122)
[X.sub.ijk]
Seeding rate (kg/ha) 0.002 (***) 0.002 (***)
(0.000) (0.001)
Age of seedlings -0.004 0.006
(days) (0.003) (0.012)
Certified (1 = yes) 0.100 (***) 0.081
(0.030) (0.071)
Area cultivated (ha) -0.420 (***) 0.154
(0.086) (0.172)
[[bar.X].sub.ij]
Mean seeding rate -0.001
(kg/ha) (0.001)
Mean age of -0.012
seedlings (days) (0.014)
Mean certified seed 0.011
(1 = yes) (0.080)
Mean area cultivated -0.735 (***)
(ha) (0.198)
[I.sub.ij]
Pesticides (1 = yes) 0.082 (**) -0.169
(0.041) (0.129)
Organic fertilizer 0.000 -0.000
(kg/ha)
(0.000) (0.000)
Chemical fertilizer 0.001 (***) 0.000
(kg/ha) (0.000) (0.001)
Labour (person- 0.000 0.000
days/ha) (0.000) (0.000)
[[bar.I].sub.i]
Mean pesticides 0.270 (**)
(1 = yes) (0.132)
Mean organic 0.000
fertilizer (kg/ha) (0.000)
Mean chemical 0.001 (*)
fertilizer (kg/ha) (0.001)
Mean labour -0.000
(person-days/ha) (0.000)
[P.sub.ij]
Slope (1 = yes) -0.080 (***) -0.042
(0.027) (0.101)
Irrigated (1 = yes) 0.114 (***) 0.053
(0.042) (0.100)
Prone to drought 0.009 -0.143 (*)
(1 = yes) (0.030) (0.083)
[[bar.P].sub.i]
Mean slope (1 = yes) -0.039
(0.105)
Mean irrigation 0.050
(1 = yes) (0.112)
Mean prone to 0.182 (**)
drought (1 = yes) (0.087)
[H.sub.i]
Elevation (masl) -0.000 (***) -0.000 (***)
(0.000) (0.000)
Sex (1 = female) -0.019 -0.020
(0.029) (0.029)
Literate (1 = yes) -0.001 -0.004
(0.028) (0.028)
SPG village 0.095 (***) 0.097 (***)
(1 = yes) (0.030) (0.030)
Constant 7.863 (***) 7.926 (***)
(0.122) (0.132)
Number. of 1,479 1,479
observations
Number of plots 1,126 1,126
(2) (4)
Variables Yield variance: Plot Yield variance: Plot
RE CRE
[T.sub.ijk]
Variety type
Old MV -0.042 -0.062
(0.134) (0.151)
New MV (not -0.172 (**) -0.103
STRV) (0.074) (0.086)
STRV -0.238 (***) -0.455 (***)
(0.085) (0.174)
Hybrid -0.162 (*) -0.219 (*)
(0.091) (0.112)
[[bar.T].sub.ij]
Mean old MV 0.054
(0.243)
Mean new MV -0.088
(not STRV)
(0.140)
Mean STRV 0.336
(0.220)
Mean hybrid 0.086
(0.166)
[X.sub.ijk]
Seeding rate (kg/ha) 0.000 0.001
(0.000) (0.001)
Age of seedlings -0.001 -0.029
(days) (0.003) (0.020)
Certified (1 = yes) -0.055 -0.063
(0.039) (0.078)
Area cultivated (ha) -0.401 (**) -0.482 (**)
(0.165) (0.227)
[[bar.X].sub.ij]
Mean seeding rate -0.001
(kg/ha) (0.001)
Mean age of 0.032
seedlings (days) (0.021)
Mean certified seed 0.020
(1 = yes) (0.094)
Mean area cultivated 0.139
(ha) (0.302)
[I.sub.ij]
Pesticides (1 = yes) -0.016 -0.076
(0.047) (0.107)
Organic fertilizer 0.000 0.000
(kg/ha)
(0.000) (0.000)
Chemical fertilizer -0.001 (*) 0.000
(kg/ha) (0.000) (0.001)
Labour (person- 0.000 -0.000
days/ha) (0.000) (0.000)
[[bar.I].sub.i]
Mean pesticides 0.060
(1 = yes) (0.111)
Mean organic -0.000
fertilizer (kg/ha) (0.000)
Mean chemical -0.001 (*)
fertilizer (kg/ha) (0.001)
Mean labour 0.000
(person-days/ha) (0.000)
[P.sub.ij]
Slope (1 = yes) 0.118 (***) -0.073
(0.042) (0.089)
Irrigated (1 = yes) 0.007 0.129
(0.066) (0.130)
Prone to drought 0.037 0.156
(1 = yes) (0.066) (0.102)
[[bar.P].sub.i]
Mean slope (1 = yes) 0.202 (**)
(0.100)
Mean irrigation -0.104
(1 = yes) (0.161)
Mean prone to -0.141
drought (1 = yes) (0.115)
[H.sub.i]
Elevation (masl) 0.000 -0.000
(0.000) (0.000)
Sex (1 = female) -0.029 -0.030
(0.026) (0.024)
Literate (1 = yes) 0.076 0.082 (*)
(0.048) (0.047)
SPG village -0.050 -0.067 (**)
(1 = yes) (0.035) (0.031)
Constant 0.339 (**) 0.221
(0.149) (0.194)
Number. of 1,479 1,479
observations
Number of plots 1,126 1,126
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
per cent, respectively. Standard errors are given in parentheses. All
standard errors are robust to heteroscedasticity. The number of
observations differs due to missing observations.
Table A6. Household RE and CRE estimates of the impact of the
cultivation of STRVs on input use and legume and vegetable cultivation
for plots with only one variety
(1) (2) (3)
Variables Labour Labour CRE Organic
RE fertilizer RE
[T.sub.ij]
Variety type
Old MV 30.877 26.882 -1,256.569
(66.957) (111.044) (1,540.400)
New MV 57.152 73.645 -2,444.230
(not STRV) (70.244) (116.365) (1,486.828)
STRV 40.656 -10.849 -440.950
(67.377) (109.896) (1,527.441)
Hybrid 7.720 46.080 -22.171
(66.519) (108.374) (1,411.773)
[[bar.T].sub.i]
Mean old MV 13.753
(154.967)
Mean new MV -44.007
(not STRV)
(153.586)
Mean STRV 105.792
(157.389)
Mean hybrid -86.142
(142.589)
[P.sub.ij]
Slope (1 = yes) -10.659 31.935 1,260.885
(39.573) (61.126) (802.975)
Irrigated 88.066 -140.832 -1,915.260 (*)
(1 = yes)
(53.570) (93.696) (1,002.176)
Prone to drought 117.279 (***) -96.827 -990.124
(1 = yes)
(35.380) (63.675) (695.195)
Plot size (ha) -830.966 -1,460.330 -10,632.080
(***) (***) (***)
(124.068) (255.708) (2,100.193)
[[bar.P].sub.i]
Mean slope -58.375
(1 = yes) (76.758)
Mean irrigated 395.368 (***)
(1 = yes) (105.985)
Mean prone to 297.112 (***)
drought (1 = yes) (79.775)
Mean plot size 834.570 (***)
(ha) (267.431)
[H.sub.i]
Elevation (masl) -0.260 (***) -0.251 (***) -11.258 (***)
(0.082) (0.088) (1.781)
Sex (1 = female) -29.943 -32.191 370.595
(50.344) (50.810) (970.327)
Literate (1 = yes) -61.373 -56.305 947.672
(45.673) (43.778) (843.807)
SPG village (1 = -113.924 -110.223 (**) 646.715
yes) (**)
(49.405) (50.623) (1,056.586)
Constant 872.826 649.452 (***) 19,438.718
(***) (***)
(113.440) (123.727) (2,696.022)
Number of 940 940 937
observations
Number of 683 683 681
households
(4) (5) (6)
Variables Organic Chemical Chemical
fertilizer fertilizer fertilizer
CRE RE CRE
[T.sub.ij]
Variety type
Old MV 269.031 38.495 (***) 30.916 (*)
(2,258.184) (11.861) (16.801)
New MV -836.606 34.294 (***) 39.516 (***)
(not STRV) (2,022.753) (10.447) (13.630)
STRV 930.113 22.605 (**) 43.998 (***)
(2,086.561) (10.119) (14.147)
Hybrid 2,494.262 25.711 (**) 25.687 (*)
(1,806.849) (10.683) (15.326)
[[bar.T].sub.i]
Mean old MV -3,586.286 17.937
(2,951.043) (26.041)
Mean new MV -3,983.109 0.582
(not STRV)
(2,709.866) (24.326)
Mean STRV -3,985.996 -32.592
(2,786.548) (24.172)
Mean hybrid -5,415.133 (**) 8.316
(2,568.400) (26.112)
[P.sub.ij]
Slope (1 = yes) 1,629.572 4.923 4.822
(1,507.381) (8.461) (25.990)
Irrigated -3,584.379 (**) 4.376 4.830
(1 = yes)
(1,486.380) (13.746) (22.404)
Prone to drought -400.561 26.808 (***) -13.285
(1 = yes)
(1,146.337) (7.382) (12.815)
Plot size (ha) -15,744.479 -108.904 -106.879 (***)
(***) (***)
(4,293.996) (17.725) (37.421)
[[bar.P].sub.i]
Mean slope -298.696 -3.355
(1 = yes) (1,795.874) (27.785)
Mean irrigated 3,948.487 (**) -10.660
(1 = yes) (1,898.217) (26.773)
Mean prone to -1,115.924 61.661 (***)
drought (1 = yes) (1,467.443) (16.597)
Mean plot size 6,840.822 -5.278
(ha) (4,583.382) (40.928)
[H.sub.i]
Elevation (masl) -11.575 (***) -0.072 (***) -0.069 (***)
(1.842) (0.020) (0.020)
Sex (1 = female) 363.292 5.384 5.114
(967.969) (10.634) (10.224)
Literate (1 = yes) 1,067.676 11.518 11.423
(835.018) (8.257) (8.070)
SPG village (1 = 627.218 -16.649 (*) -13.313
yes)
(1,067.159) (9.394) (9.473)
Constant 19,693.789 139.881 134.872 (***)
(***) (***)
(3,106.307) (24.224) (27.036)
Number of 937 935 935
observations
Number of 681 680 680
households
(7) (9)
Variables Legumes/ Legumes/
vegetables vegetables
RE CRE
[T.sub.ij]
Variety type
Old MV 0.020 0.019
(0.061) (0.097)
New MV 0.016 0.005
(not STRV) (0.058) (0.088)
STRV 0.044 0.020
(0.060) (0.091)
Hybrid 0.045 0.017
(0.058) (0.087)
[[bar.T].sub.i]
Mean old MV -0.006
(0.127)
Mean new MV 0.007
(not STRV)
(0.119)
Mean STRV 0.023
(0.123)
Mean hybrid 0.037
(0.118)
[P.sub.ij]
Slope (1 = yes) 0.169 (***) 0.128
(0.038) (0.095)
Irrigated -0.010 -0.025
(1 = yes)
(0.044) (0.073)
Prone to drought -0.053 (*) -0.009
(1 = yes)
(0.032) (0.063)
Plot size (ha) 0.103 0.173
(0.088) (0.192)
[[bar.P].sub.i]
Mean slope 0.053
(1 = yes) (0.103)
Mean irrigated 0.030
(1 = yes) (0.093)
Mean prone to -0.060
drought (1 = yes) (0.073)
Mean plot size -0.087
(ha) (0.215)
[H.sub.i]
Elevation (masl) 0.000 0.000
(0.00) (0.000)
Sex (1 = female) -0.125 -0.122
(***) (***)
(0.042) (0.042)
Literate (1 = yes) 0.118 (***) 0.118 (***)
(0.037) (0.037)
SPG village (1 = 0.150 (***) 0.147 (***)
yes)
(0.042) (0.042)
Constant
Number of 940 940
observations
Number of 683 683
households
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
er cent, respectively. Standard errors are given in parentheses. All
standard errors are robust to heteroscedasticity, except Logit FE
results which do not allow robust standard errors. The number of
observations differs due to missing observations. Logit results are
presented as marginal effects (delta method standard errors).
Table A7. Household RE and CRE estimates of the impact of the
cultivation of STRVs on early-season input use for plots with only one
variety
(1) (2) (3)
Variables Early- Early- Land
season season preparation
chemical chemical labour
fertilizer fertilizer (detailed)
(detailed) (detailed) Household
Household Household RE
RE CRE
[T.sub.ij]
Variety type
Old MV 29.517 (**) 21.107 25.007 (*)
(12.794) (19.563) (14.044)
New MV 16.797 17.952 19.560
(not STRV) (10.711) (13.359) (13.238)
STRV 19.430 (**) 28.961 (***) 27.158 (**)
(9.820) (11.234) (13.283)
Hybrid 15.309 10.900 19.164
(10.268) (10.629) (12.242)
[[bar.T].sub.i]
Mean MV 15.204
(24.807)
Mean new 2.983
MV
(not STRV)
(22.088)
Mean STRV -12.516
(20.542)
Mean hybrid 11.633
(21.627)
[P.sub.ij]
Slope 7.121 2.837 21.300 (**)
(1 = yes)
(6.742) (20.739) (9.117)
Irrigated 11.204 26.529 -24.081 (*)
(1 = yes) (11.464) (18.988) (13.503)
Prone to 17.026 (**) 9.084 -5.886
drought (6.933) (11.094) (9.209)
(1 = yes)
Plot size -63.559 -86.305 -154.357
(ha) (***) (***) (***)
(14.625) (33.314) (38.526)
[[bar.P].sub.i]
Mean slope 5.206
(1 = yes) (21.889)
Mean -32.333
irrigated (24.995)
(1 = yes)
Mean prone 12.708
to drought (13.963)
(1 = yes)
Mean plot 31.266
size (ha) (34.875)
[H.sub.i]
Elevation -0.018 -0.020 0.028
(masl)
(0.027) (0.026) (0.024)
Sex 13.788 (*) 15.042 (*) -28.555 (**)
(1 = female)
(8.000) (7.945) (11.145)
Literate 9.923 10.947 2.769
(1 = yes)
(7.210) (7.237) (10.897)
SPG village -9.319 -8.357 10.261
(1 = yes) (8.119) (8.303) (14.136)
Constant 38.619 (*) 44.216 (*) 131.619
(***)
(22.951) (25.583) (29.810)
Number of 419 419 420
observations
Number of 308 308 308
households
(4) (5) (6)
Variables Land Male land Male land
preparation preparation preparation
labour labour labour
(detailed) (detailed) (detailed)
Household Household Household
CRE RE CRE
[T.sub.ij]
Variety type
Old MV 23.041 15.961 (**) 22.937 (**)
(22.846) (6.797) (11.694)
New MV 19.305 15.733 (**) 26.725 (**)
(not STRV) (23.210) (6.618) (13.198)
STRV 40.986 (**) 18.263 (***) 33.119 (***)
(19.128) (6.851) (10.673)
Hybrid 33.894 (*) 11.642 (**) 23.061 (**)
(19.948) (5.795) (11.096)
[[bar.T].sub.i]
Mean MV -2.853 -13.590
(31.148) (17.032)
Mean new -7.531 -20.299
MV
(not STRV)
(31.175) (17.899)
Mean STRV -30.654 -26.699
(29.975) (16.941)
Mean hybrid -27.413 -19.850
(28.413) (16.135)
[P.sub.ij]
Slope -31.612 (*) 12.857 (***) -11.810
(1 = yes)
(18.634) (4.950) (10.892)
Irrigated -7.940 -9.212 -4.514
(1 = yes) (16.955) (7.193) (10.744)
Prone to 4.901 -3.774 -2.594
drought (17.699) (4.994) (11.184)
(1 = yes)
Plot size -232.552 -75.115 (***) -112.649
(ha) (***) (***)
(71.986) (19.957) (37.297)
[[bar.P].sub.i]
Mean slope 63.870 (***) 29.491 (**)
(1 = yes) (20.482) (11.908)
Mean -28.538 -7.783
irrigated (25.644) (14.049)
(1 = yes)
Mean prone -14.306 -1.855
to drought (20.352) (12.744)
(1 = yes)
Mean plot 95.675 45.131
size (ha) (80.026) (40.921)
[H.sub.i]
Elevation 0.020 0.022 (*) 0.017
(masl)
(0.025) (0.012) (0.013)
Sex -25.600 (**) -18.684 (***) -17.567 (***)
(1 = female)
(11.090) (5.492) (5.381)
Literate 4.744 8.817 10.008 (*)
(1 = yes)
(10.863) (5.478) (5.508)
SPG village 11.611 3.917 4.245
(1 = yes) (14.352) (7.809) (7.806)
Constant 144.179 48.177 (***) 56.447 (***)
(***)
(37.172) (15.547) (19.244)
Number of 420 420 420
observations
Number of 308 308 308
households
(7) (8)
Variables Female Female
land land
preparation preparation
labour labour
(detailed) (detailed)
Household Household
RE CRE
[T.sub.ij]
Variety type
Old MV 8.960 1.432
(8.650) (12.493)
New MV 3.427 -6.460
(not STRV) (7.453) (11.419)
STRV 8.343 8.268
(7.646) (9.890)
Hybrid 7.364 11.364
(7.519) (10.173)
[[bar.T].sub.i]
Mean MV 9.703
(16.716)
Mean new 12.354
MV
(not STRV)
(15.440)
Mean STRV -3.624
(14.994)
Mean hybrid -7.711
(14.403)
[P.sub.ij]
Slope 9.028 (*) -18.698 (*)
(1 = yes)
(4.970) (9.663)
Irrigated -15.772 (*) -4.935
(1 = yes) (8.185) (8.997)
Prone to -2.941 5.852
drought (5.628) (11.186)
(1 = yes)
Plot size -78.576 (***) -119.285
(ha) (***)
(20.408) (44.750)
[[bar.P].sub.i]
Mean slope 33.368 (***)
(1 = yes) (11.214)
Mean -20.084
irrigated (15.982)
(1 = yes)
Mean prone -11.458
to drought (12.651)
(1 = yes)
Mean plot 49.926
size (ha) (48.146)
[H.sub.i]
Elevation 0.004 0.002
(masl)
(0.013) (0.014)
Sex -10.068 -8.384
(1 = female)
(6.365) (6.378)
Literate -5.845 -5.172
(1 = yes)
(6.687) (6.578)
SPG village 6.474 7.467
(1 = yes) (7.530) (7.721)
Constant 85.135 (***) 88.848 (***)
(17.819) (22.731)
Number of 420 420
observations
Number of 308 308
households
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
per cent, respectively. Standard errors are given in parentheses. All
standard errors are robust to heteroscedasticity. The number of
observations differs due to missing observations.
by
Kate Vaiknoras
Catherine Larochelle
Jeffrey Alwang
About the authors
Kate Vaiknoras is a Research Agricultural Economist at the USDA Economic Research Service in the Markets and Trade Economics Division (MTED). She graduated with her PhD from the Department of Agricultural and Applied Economics at Virginia Tech in 2019; she also has an MS in Agricultural and Applied Economics from Virginia Tech and a BA in Economics from Smith College. Her research examines the adoption of agricultural technologies, particularly improved seed varieties, and the impact that adoption has on household outcomes. She has extensive experience in developing and conducting household surveys.
Catherine Larochelle has been a faculty member in the Department of Agricultural and Applied Economics at Virginia Tech since 2011. She held the positions of research associate and research assistant professor and is now a tenure-track faculty since 2015. Her research focuses on assessing the impact of agricultural research and programmes on yield, poverty, food security and nutrition. She has researched and published on issues related to food demand, farm productivity, risk management strategies, factors driving and constraining technology adoption, and gender decision-making, among others. Catherine recently studied the adoption and disadoption of iron-biofortified bean varieties and the impact of their adoption on yield, bean consumption and bean sales among rural households in Rwanda. She has been extensively involved in advising graduate students and teaches courses on international agricultural development and trade and agricultural production and consumption economics at Virginia Tech.
Jeffrey Alwang is a professor in the Department of Agricultural and Applied Economics at Virginia Tech. He received his PhD from Cornell University in 1989 and has taught and conducted research on agricultural and rural development at Virginia Tech since then. His research focuses on policies to alleviate poverty in rural areas, the development of rural economies and the assessment of impacts of technologies, policies and programmes on rural residents. His recent projects have assessed impacts of new agricultural technologies (Bangladesh, China, Ecuador, Egypt, Ethiopia, Peru, Rwanda and Uganda); strategies to engage poor producers in modern value chains (Ecuador, Indonesia and Jordan); the use of information and communications technologies to promote technology adoption (Ecuador an Rwanda); and child labour, schooling and poverty (Jordan and Zimbabwe).
Acknowledgements
Funding for this research was provided by the International Fund for Agricultural Development (IFAD) and supplemented in various ways with resources from the College of Agriculture and Life Sciences, Virginia Polytechnic and State University (Virginia Tech). This work was also partially supported by the USDA National Institute of Food and Agriculture, Hatch project VA-160102. The IFAD-funded Consortium for Unfavorable Rice Environments (CURE) provided access to data and background reports that were helpful in the completion of this project. The authors especially thank Dr. David Johnson, former director of CURE for his guidance. The International Rice Research Institute (IRRI) provided logistical support and access to background data. The authors acknowledge support with data collection from iDE Nepal and faculty and students from the Institute of Agriculture and Animal Science (IAAS) in Lamjung district, Nepal. We are particularly grateful for contributions from Arun Limbu, Bal Krishna Thapa and Rakesh Kothari from iDE Nepal; Rajkumar Pandey and Rajesh Pandey from the Child Health and Environment Save Society (CHESS) Nepal; Bhaba Tripathi from IRRI; Bishnu Bilas Adhikari from IAAS; and the enumerators and drivers who helped us complete our field work. We also acknowledge contributions of seed producer group executive committee members who participated in our focus groups, and the households and village leaders who took the time to participate in our surveys. Dr. Fabrizio Bresciani and Dr. Aslihan Arslan were instrumental in designing and conducting the study. Dr. Arslan provided numerous insights throughout the study and provided insightful comments on intermediate drafts. We thank an anonymous reviewer and anonymous IFAD advisory board members for their feedback on an earlier draft of this paper. Finally, we thank Dr. George Norton (Virginia Tech) for his helpful comments and suggestions, and participants in a workshop hosted by the International Center for Tropical Agriculture in Vietnam in August 2019.
The findings and conclusions in this presentation are those of the authors and should not be construed to represent any official US Department of Agriculture (USDA) or US government determination or policy. This research was supported in part by the USDA Economic Research Service. This work was partially completed while the main author was a graduate student at Virginia Tech.
(1) A VDC is an administrative unit larger than a village but smaller than a district.
(2) Although 50 per cent of households should have received the long survey version, only 45 per cent of households did. For 43 plot-level observations across 22 households, enumerators entered at the start of the survey that they would give the long version, but households did not answer the extended survey modules. We include these as short-survey observations so that we do not lose these observations; this does not affect the statistical similarity between the short-survey and long-survey groups of observations.
(3) This spans the months of June and July.
(4) This spans October and November.
(5) Of the households that grew an STRV in 2018, 15 per cent also grew a local variety, 19 per cent grew an old improved variety, 31 per cent grew another new improved variety, and 33 per cent grew a hybrid variety. Of the plots that had an STRV, 8 per cent also had a local variety, 10 per cent had an old improved variety, 11 per cent had another new improved variety, and 13 per cent had a hybrid variety.
(6) Pesticide use is very low in the area, so we include it only as a dummy variable. Only 51 households that received the longer module applied any pesticides, and they mostly only applied one type once. Thus, the longer module did not provide much new information about pesticides.
(7) We did not convert these to an adult-equivalent measure because a negligible quantity of labour was done by children: only 0.91 per cent of plots had any child labour used on them.
(8) Because harvesting and threshing are done after yields have been realized, we also estimated this model using only land preparation and weeding labour variables as part of the set of detailed labour variables; the remaining labour variables are still insignificant.
(9) We also estimated these models using only plots that were cultivated with one type of variety (Tables A2 and A3). Results are very similar to those presented in Table 7, except that all MV and hybrid variety types have increased male land preparation labour.
Table 1. Means (standard deviations) of inputs at the plot level,
short-version vs. long-version survey responses
Variable Short-survey
responses from
plots of
households only
assigned the short
survey
Organic fertilizer (kg/ha): short-survey 8,682.94
responses compared to sum of long- (10,046.27)
survey responses
Chemical fertilizer (kg/ha): short-survey 120.30
responses compared to sum of urea, DAP (106.74)
and potash responses of long-survey
responses
Pesticides (l/ha): short-survey responses 0.41 (1.94)
compared to sum of long-survey
responses
Labour (person-days/ha): short-survey 619.06
responses compared to sum of weeding, (575.95)
land preparation, harvesting and
threshing responses of long-survey
responses
Number of observations 660
Variable Short-survey
responses from
plots of
households
assigned the
long survey
Organic fertilizer (kg/ha): short-survey 8,635.81
responses compared to sum of long- (9,720.60)
survey responses
Chemical fertilizer (kg/ha): short-survey 121.27
responses compared to sum of urea, DAP (113.11)
and potash responses of long-survey
responses
Pesticides (l/ha): short-survey responses 0.38 (2.10)
compared to sum of long-survey
responses
Labour (person-days/ha): short-survey 576.67
responses compared to sum of weeding, (488.53)
land preparation, harvesting and
threshing responses of long-survey
responses
Number of observations 544
Variable Long-survey
responses from
plots of
households
assigned the
long survey
Organic fertilizer (kg/ha): short-survey 7,974.79
responses compared to sum of long- (10,627.71)
survey responses
Chemical fertilizer (kg/ha): short-survey 121.87
responses compared to sum of urea, DAP (112.77)
and potash responses of long-survey
responses
Pesticides (l/ha): short-survey responses 0.34 (1.65)
compared to sum of long-survey
responses
Labour (person-days/ha): short-survey 323.59
responses compared to sum of weeding, (193.98) (***)
land preparation, harvesting and
threshing responses of long-survey
responses
Number of observations 544
Note: (*/**/***) denotes statistical significance at 10, 5 and 1 per
cent, respectively. Organic fertilizer estimate is based on 1,200 total
observations, while chemical fertilizer is based on 1,199 observations
due to missing responses. We also tested differences in short-survey
responses between households that were assigned the short vs. the long
survey and found that they did not differ from one another.
Table 2. Distribution of rice seed types and mean (standard deviation)
of characteristics of different rice variety types, 2018
Variety Distribution Most Mult. Seeding Season
type of varieties, prevalent ratio rate, duration,
% (N) type on kg/ha weeks
plot
Local 14.38% 9.18% 51.26 89.99 19.43
(235) (51.29) (79.29) (2.52)
(***) (***)
Old MV 18.79% 20.70% 52.98 121.44 18.66
(307) (47.85) (70.80) (1.72)
(***) (***) (***)
New 23.75% 27.21% 72.91 97.63 18.38
MV (388) (63.56) (69.80) (2.52)
(***)
STRV (1) 19.77% 21.04% 73.30 93.40 17.72
(323) (75.46) (75.78) (2.52)
Hybrid 23.32% 21.87% 315.14 22.09 17.96
(381) (135.71) (29.75) (2.14)
(***) (***)
Total 100% 100% 121.77 82.35 18.34
(1,634) (135.51) (73.35) (2.22)
Variety Planting Harvest Grown on Grown on
type week week irrigated plot prone
(from first (from first plot to drought
week of week of (1= yes) (1 = yes)
year) year)
Local 11.19 30.62 0.89 0.52
(2.15) (1.51) (0.31) (0.03)
(**) (***) (***) (**)
Old MV 11.05 29.71 0.91 0.44
(11.64) (1.38) (0.29) (0.50)
(***) (***) (***) (***)
New 11.07 29.46 0.92 0.44
MV 1.91) (1.43) (0.27) (0.50)
(***) (***) (***)
STRV (1) 11.64 29.36 0.73 0.65
(2.19) (1.52) (0.45) (0.48)
Hybrid 11.43 29.36 0.88 0.50
(1.88) (1.52) (0.32) (0.50)
(***) (***)
Total 11.28 29.62 0.87 0.51
(1.21) (1.49) (0.34) (0.50)
Note: (*)/(**)/(***) denotes that the mean is different from that of
STRVs at a 10, 5 or 1 per cent level of significance.
(1) Of the nine STRVs grown by farmers in our sample, eight were
drought-tolerant, and one was submergence-tolerant, Swarna sub1. Swarna
sub1 made up 6 per cent of the STRV observations in our sample; 90 per
cent of observations for this variety were planted on irrigated plots,
and 30 per cent on plots prone to drought.
Table 3. Means (standard deviations) of plot-level outcome variables
and characteristics
Plots on which Plots on which non-
Variable STRVs are the most STRVs are the most
prevalent type prevalent type
Labour (person-days/ha) 626.81 (580.66) 594.69 (527.58)
Organic fertilizer (kg/ha) 9,266.12 (9,940.20) 8,515.27 (9,903.92)
Chemical fertilizer (kg/ha) 116.07 (99.44) 122.04 (112.38)
Pesticides (l/ha) 0.49 (2.70) 0.35 (1.68)
Grew legumes and/or 0.35 (0.48) 0.32 (0.46)
vegetables in monsoon
season (1= yes) 0.79 (0.41) 0.68 (0.47) (***)
Slope (1 = yes) 0.08 (0.28) 0.03 (0.18) (***)
Suffered from drought, 2018 0.21 (0.41) 0.09 (0.28) (***)
Suffered from drought, 2017 252 946
Number of observations
Note: (*/**/***) denotes statistical significance at 10, 5 and 1 per
cent, respectively.
Table 4. Explanatory variables used in analysis
Variable Description
[S.sub.ijk] Variety-level
variables
Seeding rate Seeding rate or density of seedlings planted
in kg/ha of variety k on plot j
Age of seedlings Age of seedlings (in days) when
transplanted to the field
Certified seed At least 50% of planting material was
certified or truthfully labelled
Area cultivated Area on plot j that variety k was grown
on in hectares
[I.sub.ij] Input
variables (l = 1) Total quantity of chemical fertilizer in
Chemical fertilizer kg/ha applied on plot
Total quantity of organic fertilizer in
Organic fertilizer kg/ha applied on plot
Total person-days of labour/ha (both unpaid
Labour and hired) applied on plot
Household applied pesticides
Pesticides to plot k (1 = yes)
[I.sub.ij] Input
variables (l = 2) Total quantity of urea in kg/ha
Urea applied on plot
Total quantity of DAP in kg/ha
DAP applied on plot
Total quantity of chemical potash
Potash in kg/ha applied on plot
Sum of organic fertilizer applications in
Organic fertilizer kg/ha applied on plot, obtained
(detailed) from detailed questioning of fertilizer
applications
Total person-days/ha of paid and unpaid
Pand preparation labour devoted to nursery
and planting preparation, land preparation and planting
labour Total person-days/ha of paid and unpaid
Weeding/pest labour devoted to weed and
control labour pest control
Total person-days/ha of paid and unpaid
Harvesting labour labour devoted to harvesting
Total person-days/ha of paid and
Threshing labour unpaid labour devoted to threshing
Household applied pesticides to
Pesticides plot (1 = yes)
[P.sub.ij] Plot
characteristics Plot is irrigated (1 = yes)
Irrigated 0 = flat; 1 = gentle, moderate
Slope or steep slope
Susceptibility of rice plots to drought, as
assessed by the farmer (0 = not
at all; 1 = somewhat; average; very;
extremely)
Prone to drought Size of plot in hectares
Plot size
[H.sub.i] Household
characteristics Elevation of the household, in metres above
Elevation sea level (masl)
Sex of head of household (1 = female)
Sex Head of household is literate (1 = yes)
Literate There is an SPG in the village
SPG village
Table 5. Coefficient (standard error) of RE and CRE estimates of the
impact of variety type and covariates on mean yield, yield variance and
growing duration
(1) (2) (3)
Variables Yield mean: Yield Season
Plot RE variance: duration:
Plot RE Plot RE
[T.sub.ijk]
Variety type
Old MV 0.208 (***) -0.014 -0.321 (*)
(0.069) (0.134) (0.176)
New MV (not 0.278 (***) -0.118 -0.646 (***)
STRV)
(0.064) (0.077) (0.194)
STRV 0.268 (***) -0.196 (**) -1.316 (***)
(0.064) (0.089) (0.222)
Hybrid 1.242 (***) -0.057 -1.068 (***)
(0.071) (0.092) (0.238)
[[bar.T].sub.ij]
Mean old MV
Mean new MV
(not STRV)
Mean STRV
Mean hybrid
[X.sub.ijk]
Seeding rate -0.008 (***) 0.001 (**) 0.001
(kg/ha)
(0.000) (0.000) (0.001)
Age of seedlings -0.005 -0.004 0.011
(days)
(0.004) (0.004) (0.019)
Certified (1 = yes) 0.169 (***) -0.012 0.089
(0.038) (0.048) (0.134)
Area cultivated (ha) -0.259 (*) -0.124 0.780 (**)
(0.133) (0.209) (0.358)
[[bar.X].sub.ij]
Mean seeding
rate (kg/ha)
Mean age of
seedlings (days)
Mean certified
seed (1 = yes)
Mean area
cultivated (ha)
[I.sub.ij]
Pesticides (1 = yes) 0.085 (*) -0.041 -0.305
(0.045) (0.052) (0.219)
Organic fertilizer 0.000 0.000 0.000 (*)
(kg/ha)
(0.000) (0.000) (0.000)
Chemical fertilizer 0.001 (***) -0.001 (**) -0.002 (***)
(kg/ha)
(0.000) (0.000) (0.001)
Labour (person- -0.000 0.000 0.000 (**)
days/ha)
(0.000) (0.000) (0.000)
[[bar.I].sub.i]
Mean pesticides
(1 = yes)
Mean organic
fert (kg/ha)
Mean chemical
fert (kg/ha)
Mean labour
(person-days/ha)
[P.sub.ij]
Slope (1 = yes) -0.075 (**) 0.097 (**) 0.753 (***)
(0.033) (0.048) (0.145)
Irrigated (1 = yes) 0.118 (**) 0.016 0.550 (***)
(0.048) (0.074) (0.209)
Prone to drought -0.006 0.053 -0.002
(1 = yes)
(0.034) (0.070) (0.135)
[[bar.P].sub.i]
Mean slope
(1 = yes)
Mean irrigation
(1 = yes)
Mean prone to
drought (1 =
yes)
[H.sub.i]
Elevation (masl) -0.000 (***) 0.000 0.001 (***)
(0.000) (0.000) (0.000)
Sex (1 = female) -0.032 -0.028 -0.002
(0.034) (0.033) (0.152)
Literate (1 = yes) 0.009 0.063 -0.393 (***)
(0.034) (0.054) (0.140)
SPG village 0.156 (***) -0.010 0.070
(1 = yes)
(0.038) (0.042) (0.168)
Constant 4.487 (***) 0.196 17.073 (***)
(0.147) (0.170) (0.639)
Number of 1,467 1,467 1,443
observations
Number of plots 1,117 1,117 1,106
(4) (5) (6)
Variables Yield Yield Season
mean: Plot variance: duration:
CRE Plot CRE Plot CRE
[T.sub.ijk]
Variety type
Old MV 0.151 -0.093 -0.165
(0.122) (0.145) (0.216)
New MV (not 0.159 -0.150 (*) -0.637 (***)
STRV)
(0.102) (0.088) (0.234)
STRV 0.297 (**) -0.450 (***) -1.264 (***)
(0.120) (0.174) (0.317)
Hybrid 1.190 (***) -0.143 -1.513 (***)
(0.115) (0.132) (0.346)
[[bar.T].sub.ij]
Mean old MV 0.086 0.134 -0.389
(0.136) (0.248) (0.362)
Mean new MV 0.178 0.053 -0.121
(not STRV)
(0.121) (0.154) (0.380)
Mean STRV -0.018 0.385 (*) -0.144
(0.138) (0.226) (0.455)
Mean hybrid 0.093 0.132 0.562
(0.143) (0.194) (0.488)
[X.sub.ijk]
Seeding rate -0.007 (***) 0.002 -0.003
(kg/ha)
(0.001) (0.001) (0.002)
Age of seedlings -0.011 -0.028 -0.036
(days)
(0.014) (0.018) (0.059)
Certified (1 = yes) 0.116 0.019 0.218
(0.087) (0.101) (0.225)
Area cultivated (ha) 0.390 (**) -0.114 1.186 (*)
(0.199) (0.264) (0.641)
[[bar.X].sub.ij]
Mean seeding -0.000 -0.001 0.006 (**)
rate (kg/ha)
(0.001) (0.001) (0.002)
Mean age of 0.006 0.029 0.057
seedlings (days)
(0.014) (0.019) (0.061)
Mean certified 0.054 -0.034 -0.210
seed (1 = yes)
(0.100) (0.118) (0.282)
Mean area -0.812 (***) 0.049 -0.663
cultivated (ha)
(0.259) (0.377) (0.784)
[I.sub.ij]
Pesticides (1 = yes) -0.170 -0.115 -0.058
(0.155) (0.118) (0.630)
Organic fertilizer -0.000 0.000 -0.000
(kg/ha)
(0.000) (0.000) (0.000)
Chemical fertilizer 0.001 0.000 0.000
(kg/ha)
(0.000) (0.000) (0.002)
Labour (person- 0.000 0.000 0.000
days/ha)
(0.000) (0.000) (0.000)
[[bar.I].sub.i] 0.275 (*) 0.070 -0.244
Mean pesticides
(1 = yes)
(0.158) (0.123) (0.658)
Mean organic 0.000 -0.000 0.000
fert (kg/ha)
(0.000) (0.000) (0.000)
Mean chemical 0.000 -0.001 (*) -0.003
fert (kg/ha)
(0.000) (0.001) (0.002)
Mean labour -0.000 0.000 -0.000
(person-days/ha)
(0.000) (0.000) (0.000)
[P.sub.ij]
Slope (1 = yes) -0.104 -0.022 -0.413
(0.099) (0.094) (0.549)
Irrigated (1 = yes) 0.097 -0.006 0.284
(0.099) (0.080) (0.502)
Prone to drought -0.114 0.068 -0.595
(1 = yes)
(0.092) (0.099) (0.411)
[[bar.P].sub.i]
Mean slope 0.031 0.132 1.231 (**)
(1 = yes)
(0.105) (0.107) (0.572)
Mean irrigation 0.004 0.081 0.283
(1 = yes)
(0.112) (0.116) (0.544)
Mean prone to 0.131 -0.018 0.682
drought (1 =
yes)
(0.099) (0.115) (0.431)
[H.sub.i]
Elevation (masl) -0.000 (***) 0.000 0.001 (***)
(0.000) (0.000) (0.000)
Sex (1 = female) -0.033 -0.028 0.014
(0.034) (0.031) (0.152)
Literate (1 = yes) 0.006 0.068 -0.376 (***)
(0.034) (0.053) (0.140)
SPG village 0.154 (***) -0.034 0.097
(1 = yes)
(0.037) (0.038) (0.169)
Constant 4.485 (***) 0.011 16.850 (***)
(0.161) (0.227) (0.717)
Number of 1,467 1,467 1,443
observations
Number of plots 1,117 1,117 1,106
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
per cent, respectively. All standard errors are robust to
heteroscedasticity.
Table 6. Coefficient (standard error) of RE estimates of the impact of
variety type and covariates on mean yield, yield variance and growing
duration, comparing basic with detailed input variables for households
that received the long survey version
(1) (2) (3)
Yield mean: Yield Season
Plot RE variance: duration: Plot
Plot RE RE
Variables l = 1 l = 1 l = 1
[T.sub.ijk]
Variety type
Old MP 0.188 (*) -0.213 (**) -0.477 (**)
(0.108) (0.103) (0.242)
New MP (not 0.263 (**) -0.134 -0.646 (**)
STRV)
(0.106) (0.106) (0.301)
STRV 0.226 (**) -0.246 (**) -1.334 (***)
(0.104) (0.115) (0.313)
Hybrid 1.190 (***) -0.019 -1.019 (***)
(0.130) (0.134) (0.343)
[X.sub.ijk]
Seeding rate -0.008 (***) 0.002 (*) 0.001
(kg/ha)
(0.001) (0.001) (0.002)
Age of seedlings -0.015 (***) -0.003 0.000
(days)
(0.005) (0.006) (0.028)
Certified seed 0.131 (**) 0.046 0.004
(1 = yes)
(0.062) (0.058) (0.203)
Area cultivated -0.363 (**) -0.088 0.751
(ha)
(0.179) (0.143) (0.625)
[I.sub.ijl=1]
Pesticides 0.138 (**) -0.108 (***) -0.299
(1 = yes)
(0.062) (0.041) (0.407)
Organic fertilizer -0.000 0.000 0.000
(kg/ha)
(0.000) (0.000) (0.000)
Chemical fertilizer 0.001 (***) -0.001 (**) -0.003 (***)
(kg/ha)
(0.000) (0.000) (0.001)
Labour 0.000 0.000 0.000
(person days/ha)
(0.000) (0.000) (0.000)
[I.sub.ijl=2]
Organic fertilizer
from
detailed survey
(kg/ha)
Urea (kg/ha)
DAP (kg ha)
Potash (kg ha)
Land preparation
labour (person-
days/ha)
Weeding labour
(person-days/ha)
Harvesting labour
(person-days ha)
Threshing labour
(person-days/ha)
[P.sub.ij] -0.073 0.067 0.666 (***)
Slope (1 = yes) (0.047) (0.048) (0.215)
0.113 0.058 0.470
Irrigated (1 = yes) (0.070) (0.077) (0.312)
-0.036 0.087 -0.139
Prone to drought
(1 = yes) (0.048) (0.071) (0.205)
[H.sub.i] -0.000 (**) 0.000 (**) 0.001 (**)
Elevation (masl) (0.000) (0.000) (0.001)
-0.089 (**) -0.079 (*) 0.095
Sex (1 = female) (0.046) (0.045) (0.235)
-0.031 -0.020 -0.376 (*)
Literate (1 = yes) (0.048) (0.081) (0.210)
0.203 (***) 0.041 -0.021
SPG village
(1 = yes) (0.058) (0.053) (0.229)
4.877 (***) 0.038 17.745 (***)
Constant (0.230) (0.256) (0.969)
644 644 636
Number of
observations 499 499 497
Number of plots
(4) (5) (6)
Yield Yield Season
mean: Plot variance: duration:
RE Plot RE Plot RE
Variables l = 2 l = 2 l = 2
[T.sub.ijk]
Variety type
Old MP 0.195 (*) -0.206 (**) -0.466 (*)
(0.106) (0.100) (0.240)
New MP (not 0.275 (***) -0.135 -0.656 (**)
STRV)
(0.105) (0.105) (0.304)
STRV 0.227 (**) -0.244 (**) -1.325 (***)
(0.103) (0.117) (0.320)
Hybrid 1.199 (***) -0.016 -1.069 (***)
(0.130) (0.137) (0.350)
[X.sub.ijk]
Seeding rate -0.007 (***) 0.002 (*) 0.001
(kg/ha)
(0.001) (0.001) (0.002)
Age of seedlings -0.015 (***) -0.003 -0.004
(days)
(0.005) (0.006) (0.028)
Certified seed 0.126 (**) 0.054 0.068
(1 = yes)
(0.061) (0.058) (0.205)
Area cultivated -0.431 (**) -0.127 0.382
(ha)
(0.190) (0.144) (0.667)
[I.sub.ijl=1]
Pesticides 0.128 (**) -0.114 (***) -0.324
(1 = yes)
(0.062) (0.042) (0.402)
Organic fertilizer
(kg/ha)
Chemical fertilizer
(kg/ha)
Labour
(person days/ha)
[I.sub.ijl=2]
Organic fertilizer -0.000 0.000 -0.000
from
detailed survey
(kg/ha) (0.000) (0.000) (0.000)
0.001 (***) -0.001 (***) -0.002
Urea (kg/ha) (0.000) (0.000) (0.002)
0.001 (**) -0.001 (*) -0.002
DAP (kg ha) (0.000) (0.000) (0.002)
0.001 -0.001 -0.006
Potash (kg ha) (0.002) (0.002) (0.007)
-0.000 -0.000 0.001
Land preparation (0.000) (0.000) (0.001)
labour (person-
days/ha) -0.000 -0.000 -0.001
Weeding labour (0.000) (0.001) (0.002)
(person-days/ha) 0.000 0.000 0.005 (*)
Harvesting labour (0.001) (0.001) (0.003)
(person-days ha) -0.000 -0.000 -0.008 (**)
Threshing labour (0.001) (0.001) (0.003)
(person-days/ha)
[P.sub.ij] -0.055 0.068 0.660 (***)
Slope (1 = yes) (0.050) (0.051) (0.220)
0.111 0.076 0.544 (*)
Irrigated (1 = yes) (0.070) (0.088) (0.313)
-0.034 0.102 -0.073
Prone to drought
(1 = yes) (0.051) (0.080) (0.211)
[H.sub.i] -0.000 (**) 0.000 (**) 0.001
Elevation (masl) (0.000) (0.000) (0.001)
-0.091 (**) -0.092 (**) 0.097
Sex (1 = female) (0.046) (0.045) (0.227)
-0.027 -0.021 -0.330
Literate (1 = yes) (0.049) (0.085) (0.211)
0.194 (***) 0.040 -0.025
SPG village
(1 = yes) (0.057) (0.047) (0.221)
4.927 (***) 0.072 18.234 (***)
Constant (0.233) (0.250) (0.943)
645 645 637
Number of
observations 500 500 498
Number of plots
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
per cent, respectively. All standard errors are robust to
heteroscedasticity. Differences in sample sizes are due to missing
observations. Pesticides is listed as being under [I.sub.ijl=1], but it
is also in [I.sub.ijl=2].
Table 7. Household RE and CRE estimates of the impact of the
cultivation of STRVs on input use and legume and vegetable cultivation
(1) (2)
Variables Labour Labour
RE CRE
[T.sub.ij]
Variety type
Old MV 21.338 20.203
(56.805) (98.829)
New MV 25.450 61.319
(not STRV) (58.249) (105.933)
STRV 41.926 4.723
(57.113) (99.138)
Hybrid 15.237 62.813
(56.663) (98.267)
[[bar.T].sub.i]
Mean old MV 2.785
(128.506)
Mean new MV -68.537
(not STRV)
(129.992)
Mean STRV 59.706
(132.364)
Mean hybrid -100.321
(122.530)
[P.sub.ij]
Slope (1 = yes) 3.866 24.432
(34.308) (52.787)
Irrigated 105.949 (**) -114.366
(1 = yes)
(46.707) (83.289)
Prone to 98.521 (***) -71.979
drought
(1 = yes) (31.110) (53.012)
Plot size (ha) -682.358 -1,373.152
(***) (***)
(130.493) (194.281)
[[bar.P].sub.i]
Mean slope -22.184
(1 = yes) (65.934)
Mean irrigated 381.443 (***)
(1 = yes) (90.375)
Mean prone to 231.706 (***)
drought (66.564)
(1 = yes)
Mean plot size 847.748 (***)
(ha) (219.071)
[H.sub.i]
Elevation (masl) -0.264 (***) -0.264 (***)
(0.069) (0.074)
Sex (1 = female) -46.088 -37.437
(42.414) (42.860)
Literate -46.161 -44.354
(1 = yes)
(39.002) (38.117)
SPG village -133.081 -130.530
(1 = yes) (***) (***)
(40.202) (41.419)
Constant 825.517 629.283 (***)
(***)
(98.923) (102.848)
Number of 1,175 1,175
observations
Number of 872 872
households
(3) (4) (5)
Variables Organic Organic Chemical
fertilizer RE fertilizer fertilizer
CRE RE
[T.sub.ij]
Variety type
Old MV -1,339.139 239.124 31.501(***)
(1,228.495) (2,074.673) (10.240)
New MV -1,937.891 (*) -511.770 30.882(***)
(not STRV) (1,176.528) (1,799.364) (9.126)
STRV -267.888 1,180.376 21.477(**)
(1,196.429) (1,864.116) (8.888)
Hybrid 69.609 2,313.113 24.872(***)
(1,149.113) (1,610.457) (9.651)
[[bar.T].sub.i]
Mean old MV -2,800.395
(2,469.195)
Mean new MV -2,680.027
(not STRV)
(2,203.090)
Mean STRV -3,076.955
(2,264.393)
Mean hybrid -3,940.132 (*)
(2,085.001)
[P.sub.ij]
Slope (1 = yes) 1,181.207 (*) 1,137.309 4.096
(649.142) (1,280.278) (7.257)
Irrigated -1,394.989 -2,619.171 (*) 4.214
(1 = yes)
(859.735) (1,352.192) (11.755)
Prone to -457.838 421.579 22.430(***)
drought
(1 = yes) (605.570) (1,157.527) (6.270)
Plot size (ha) -7,813.246 -12,520.615 -86.087
(***) (***) (***)
(1,767.803) (3,732.828) (16.694)
[[bar.P].sub.i]
Mean slope 209.840
(1 = yes) (1,509.563)
Mean irrigated 2,709.067(*)
(1 = yes) (1,629.671)
Mean prone to -1,322.509
drought (1,366.498)
(1 = yes)
Mean plot size 5,566.940
(ha) (4,036.266)
[H.sub.i]
Elevation (masl) -10.808 (***) -11.153 (***) -0.066(***)
(1.481) (1.532) (0.017)
Sex (1 = female) 205.386 261.928 3.235
(828.074) (829.253) (8.927)
Literate 629.534 648.517 10.178
(1 = yes)
(710.925) (708.749) (7.179)
SPG village 682.478 702.412 -17.547 (**)
(1 = yes)
(879.676) (887.471) (8.077)
Constant 17,842.507 17,936.163 137.891 (***)
(***) (***)
(2,229.776) (2,458.760) (20.515)
Number of 1,172 1,172 1,169
observations
Number of 870 870 868
households
(6) (7) (9)
Variables Chemical Legumes/ Legumes/
fertilizer vegetables vegetables
CRE RE CRE
[T.sub.ij]
Variety type
Old MV 32.159(**) 0.034 0.014
(15.523) (0.052) (0.089)
New MV 40.725(***) 0.021 0.005
(not STRV) (12.808) (0.050) (0.079)
STRV 44.449(***) 0.046 0.038
(13.230) (0.051) (0.082)
Hybrid 29.024(**) 0.062 0.037
(14.157) (0.050) (0.080)
[[bar.T].sub.i]
Mean old MV 0.078 0.024
(21.250) (0.110)
Mean new MV -12.630 0.019
(not STRV)
(19.492) (0.102)
Mean STRV -35.208 (*) 0.003
(19.678) (0.106)
Mean hybrid -4.371 0.038
(21.232) (0.103)
[P.sub.ij]
Slope (1 = yes) 3.452 0.195(***) 0.132
(21.758) (0.032) (0.086)
Irrigated 4.383 -0.036 -0.030
(1 = yes)
(19.422) (0.039) (0.066)
Prone to -14.684 -0.047 (*) -0.011
drought
(1 = yes) (10.976) (0.028) (0.056)
Plot size (ha) -101.083 0.085 0.173
(***)
(28.610) (0.063) (0.166)
[[bar.P].sub.i]
Mean slope -1.707 0.074
(1 = yes) (23.443) (0.093)
Mean irrigated -5.652 -0.005
(1 = yes) (22.700) (0.083)
Mean prone to 55.921(***) -0.049
drought (14.079) (0.065)
(1 = yes)
Mean plot size 16.731 -0.101
(ha) (34.567) (0.180)
[H.sub.i]
Elevation (masl) -0.064(***) 0.000 0.000
(0.017) (0.00) (0.000)
Sex (1 = female) 3.756 -0.106 (***) -0.104 (***)
(8.700) (0.038) (0.038)
Literate 10.419 0.095(***) 0.096 (***)
(1 = yes)
(6.996) (0.033) (0.033)
SPG village -14.719 (*) 0.141(***) 0.140 (***)
(1 = yes)
(8.165) (0.038) (0.038)
Constant 135.577 (***)
(22.129)
Number of 1,169 1,175 1,175
observations
Number of 868 872 872
households
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
per cent, respectively. Standard errors are given in parentheses. All
standard errors are robust to heteroscedasticity, except Logit CRE
results which do not allow robust standard errors. The number of
observations differs due to missing observations. Logit results are
presented as marginal effects (delta method standard errors).
Table 8. Household RE and CRE estimates of the impact of the
cultivation of STRVs on early-season input use
(1) (2) (3)
Variables Early- Early- Land
season season preparation
chemical chemical labour
fertilizer fertilizer (detailed)
(detailed) (detailed) Household
Household Household RE
RE CRE
[T.sub.ij]
Variety type
Old MV 21.090 (**) 20.437 24.072 (**)
(10.741) (18.868) (11.076)
New MV 16.469 (*) 21.102 (*) 14.482
(not STRV) (8.915) (12.532) (10.472)
STRV 17.105 (**) 29.244 (**) 21.690 (**)
(8.526) (11.567) (10.882)
Hybrid 14.016 11.979 15.336
(8.969) (9.550) (10.454)
[[bar.T].sub.i]
Mean old 1.927
MV
(21.855)
Mean new -3.234
MV
(not STRV)
(17.800)
Mean STRV -13.784
(17.173)
Mean hybrid 7.590
(17.831)
[P.sub.ij]
Slope 7.574 -1.368 27.968 (***)
(1 = yes)
(5.555) (16.356) (7.462)
Irrigated ( 6.925 20.196 -32.629 (***)
1 = yes) (10.205) (17.330) (12.069)
Prone to 13.674 (**) 4.959 -5.366
drought (5.691) (9.216) (7.535)
(1 = yes)
Plot size -54.708 (***) -74.900 (***) -147.893
(ha) (***)
(11.456) (27.738) (27.394)
[[bar.P].sub.i]
Mean slope 10.702
(1 = yes) (17.372)
Mean -28.463
irrigated (22.240)
(1 = yes)
Mean prone 13.448
to drought (11.606)
(1 = yes)
Mean plot 27.838
size (ha) (29.053)
[H.sub.i]
Elevation -0.018 -0.020 0.006
(masl)
(0.021) (0.020) (0.019)
Sex 9.141 9.938 -18.930 (*)
(1 = female)
(6.910) (6.976) (9.772)
Literate 4.029 4.759 0.727
(1 = yes)
(6.224) (6.223) (8.693)
SPG village -10.478 -9.843 6.207
(1 = yes) (6.923) (7.088) (11.783)
Constant 48.640 (***) 55.028 (***) 147.361 (***)
(18.845) (20.123) (25.182)
Number of 524 524 525
observations
Number of 392 392 392
households
(4) (5) (6)
Variables Land Male land Male land
preparation preparation preparation
labour labour labour
(detailed) (detailed) (detailed)
Household Household Household
CRE RE CRE
[T.sub.ij]
Variety type
Old MV 14.382 13.825 (**) 17.318 (*)
(20.741) (5.605) (10.526)
New MV 7.936 9.479 (*) 17.284
(not STRV) (19.503) (5.455) (10.875)
STRV 29.501 (*) 12.664 (**) 24.031 (**)
(17.015) (5.945) (9.414)
Hybrid 18.439 8.659 18.220 (*)
(17.138) (5.376) (10.437)
[[bar.T].sub.i]
Mean old 11.711 -5.989
MV
(25.772) (13.994)
Mean new 6.988 -12.276
MV
(not STRV)
(24.568) (14.115)
Mean STRV -12.543 -15.713
(23.572) (13.417)
Mean hybrid -5.679 -14.294
(23.294) (13.798)
[P.sub.ij]
Slope -30.763 (*) 16.971 (***) -11.494
(1 = yes)
(17.965) (4.039) (10.675)
Irrigated ( -11.539 -10.774 (*) -1.030
1 = yes) (15.189) (6.527) (10.501)
Prone to 8.331 -3.330 1.049
drought (15.102) (4.104) (9.195)
(1 = yes)
Plot size -225.260 -72.861 (***) -112.716
(ha) (***) (***)
(63.586) (14.447) (33.046)
[[bar.P].sub.i]
Mean slope 69.100 (***) 32.545 (***)
(1 = yes) (19.370) (11.411)
Mean -34.621 -14.324
irrigated (22.338) (13.104)
(1 = yes)
Mean prone -17.664 -5.470
to drought (17.260) (10.456)
(1 = yes)
Mean plot 92.208 45.252
size (ha) (68.125) (35.126)
[H.sub.i]
Elevation 0.002 0.012 0.009
(masl)
(0.019) (0.010) (0.010)
Sex -16.554 (*) -14.899 (***) -14.145 (***)
(1 = female)
(9.701) (4.685) (4.640)
Literate 1.745 8.393 (*) 9.015 (**)
(1 = yes)
(8.615) (4.434) (4.427)
SPG village 8.036 3.081 3.974
(1 = yes) (12.032) (6.700) (6.750)
Constant 152.492 (***) 54.720 (***) 59.954 (***)
(30.090) (13.594) (15.796)
Number of 525 525 525
observations
Number of 392 392 392
households
(7) (8)
Variables Female Female
land land
preparation preparation
labour labour
(detailed) (detailed)
Household Household
RE CRE
[T.sub.ij]
Variety type
Old MV 9.613 -2.114
(6.852) (11.599)
New MV 3.856 -9.343
(not STRV) (6.072) (10.212)
STRV 7.894 5.056
(6.327) (9.100)
Hybrid 5.444 0.232
(6.242) (9.415)
[[bar.T].sub.i]
Mean old 16.668
MV
(14.001)
Mean new 19.243
MV
(not STRV)
(12.652)
Mean STRV 3.582
(12.125)
Mean hybrid 8.269
(13.252)
[P.sub.ij]
Slope 11.866 (***) -18.771 (*)
(1 = yes)
(4.047) (9.650)
Irrigated ( -23.183 (***) -11.327
1 = yes) (7.692) (8.397)
Prone to -2.953 6.662
drought (4.500) (9.720)
(1 = yes)
Plot size -74.985 (***) -112.836
(ha) (***)
(14.675) (39.397)
[[bar.P].sub.i]
Mean slope 36.277 (***)
(1 = yes) (11.203)
Mean -20.573
irrigated (13.272)
(1 = yes)
Mean prone -12.455
to drought (10.978)
(1 = yes)
Mean plot 46.447
size (ha) (41.323)
[H.sub.i]
Elevation -0.007 -0.007
(masl)
(0.011) (0.011)
Sex -4.311 -2.812
(1 = female)
(5.927) (5.968)
Literate -7.120 -6.862
(1 = yes)
(5.373) (5.307)
SPG village 3.270 4.081
(1 = yes) (6.184) (6.495)
Constant 95.072 (***) 94.373 (***)
(15.670) (18.466)
Number of 525 525
observations
Number of 392 392
households
Note: (*)/ (**)/ (***) denotes statistical significance at 10, 5 and 1
per cent, respectively. Standard errors are given in parentheses. All
standard errors are robust to heteroscedasticity. The number of
observations differs due to missing observations.
 Printer friendly
 Cite/link
 Email
 Feedback
| |
| Author: | Vaiknoras, Kate; Larochelle, Catherine; Alwang, Jeffrey |
|---|---|
| Publication: | IFAD Research Series |
| Article Type: | Survey |
| Geographic Code: | 9NEPA |
| Date: | Dec 1, 2020 |
| Words: | 18928 |
| Previous Article: | The adoption of improved agricultural technologies: A meta-analysis for Africa. |
| Next Article: | Impacts of agricultural value chain development in a mountainous region: Evidence from Nepal. |
| Topics: | |