Printer Friendly

International factory productivity gains.


A small but growing band of researchers has been applying statistical techniques to factory data to uncover policies and actions that increase productivity. This is an important issue with ramifications for both public and private management policy. While several projects are underway,(1) the published work to date is attribute either to Professors Robert H. Hayes and Kim B. Clark and their colleagues at Harvard Business School (Hayes and Clark (1985a, 1985b, 1986)) or to me and my colleagues (Schmenner and Cook (1985), Schmenner (1986, 1987, 1988a, 1988b), Schmenner and Rho (1990)).

Hayes and Clark broke new ground in a number of ways (Hayes and Clark (1985a)). They gathered individual times series data on 12 different factories of three separate companies, classified respectively as process, fab/assembly, and high tech in character. Of the twelve factories, ten were North American, one European, and one Asian. They did not pool data across factories, but instead developed monthly data from each factory for between 18 months and nine years of performance. Of particular interest to them was the development of a total factor productivity index for each factory, using that measure as the dependent variable in each regression. They also developed labor productivity measures and monthly time series on a list of 17 factory-specific managerial policies that they placed in five categories: equipment, quality, inventory, work force, and "confusion."

Most of their analysis employed linear regression, applied to the plants' time series of data, one plant at a time and one explanatory, policy variable at a time. Not a lot was found to be statistically significant. Both the high tech and process industry results were pretty lackluster, leaving most of the tales to be told by the fab/assembly plants. When they were statistically significant, the following factors appeared to be the chief influences to productivity: improved quality, measured particularly as waste rates (yield as a percent of total input materials); reduced work-in-process inventories; reduced numbers of engineering change orders (ECOs); and reduced numbers of grievances. Not all of these factors were important to each plant, but they were influential to a number of them.

Hayes and Clark did run some regressions employing all of the variables at one time, thus trying to control for the separate influences those variables might have. For the most part, the following factors had a positive influence on the productivity of the fab/assembly factories when they were examined all at the same time: more cumulative production, higher capacity utilization, less work-in-process, fewer engineering change orders (ECOs). Cumulative production(2) was included to remove the impact of the learning curve from the specific managerial variables Hayes and Clark wanted to study. Capacity utilization was investigated because the total factory productivity measure moves most dramatically with fluctuations in sales, at least in the short run, reflecting the heavy impact of capital cost, a slowly changing component of the total factor productivity measure. The two managerially most important influences, then, involved low WIP inventories and reduced "confusion," measured by the number of ECOs. Interestingly, Hayes and Clark could find no statistically significant relationship between new equipment investment and any factory productivity measure.

My own work on the topic has progressed in stages, from initial work with North Carolina factories (Schmenner and Cook (1985)), followed by a broader, more elaborate national study (Schmenner (1986)), and then on two international studies, one predominately, although not exclusively European (Schmenner (1987)), and the other exclusively Korean (Schmenner and Rho (1990)).(3) My previously published articles used the data bases separately to explore those management policies that contribute to productivity gains. For the first time, this paper consolidates the cross-sectional data from the national and international studies (561 plants in total), thereby significantly enlarging the number of observations employed and permitting some distinct tests of the data previously impossible to perform.

The tests address two hypotheses of considerable interest:

(1) It is often supposed that certain countries may be penalized or advantaged by factors such as culture, work ethic, governmental policies, regulations, or other, location-related influences. For example, do Korean factories enjoy any special factory-specific productivity advantages, other things equal, or are they simply more likely to pursue management policies that are more effective? The broad geographic spread of the consolidated data base permits tests of the influence of location-related factors on productivity.

(2) Likewise, managers frequently assert that process industries, such as food processing, textiles, chemicals, petroleum and steel, differ from non-process (more assembly-oriented) industries. The size and character of the consolidated cross-section data permit a test of this assertion as well. Do different managerial policies affect the productivity gains of process industries versus others, all other things being equal?

This paper also reports on an analysis of factory-specific times series data, using information from five plants of a major domestic auto maker (quarterly for five years). Such time series analysis, and from a single company's plants, complements the cross-sectional analysis and bolsters the conclusions. In a last section of the paper, data from five plants of the Square D Company (over a 30-month time span) are examined to call into question the use of labor efficiency as a proxy for productivity.


The cross-sectional data employed here derive from a mail survey that was originally used in the United States (1984-5). It was then used with an international sample (1986-7), heavily European, due to the International Institute for Management Development (IMD, formerly IMEDE, an international business school in Switzerland) and, in translation, in Korea (1986-7). The surveys mailed were either seven or eight pages long and covered over 220 separate questions. Table 1 reports on some of the variables and how they differ from location to location. Reference 11 lists the data asked in each survey; more information is also available from the author. The North American data base consists of 271 factory responses from a broad range of industries. There are 130 responses in the IMD sample, with most but not all of those responses concentrated in Europe. There are 160 responses to the Korean survey. [TABULAR DATA 1 OMITTED]

Numbered among the themes represented in the survey are automation and technological advance in the process, product design changes, factory complexity, quality management, elements of just-in-time manufacturing, human resource programs, methods study, factory age and equipment investment, characteristics of worker pay and organization, unionization, MRP, production planning, vendor relationships, characteristics of the plants' functional responsibilities, and management strengths and weaknesses. The data are discussed in more detain in an appendix.

The times series data employed in this paper come from the responses of five plants of a major domestic auto maker to somewhat more thorough questions about the history of productivity gain, product and process characteristics, and management programs. The responses are for the period 1984-1988, and are by quarter where such data exist. When values were only available annually, interpolations were done.

A more specialized set of data from the Square D Company is discussed in section VIII.


There are many definitions of productivity, in part because no single definition satisfies all the criteria one can apply. Theoretically, the most appealing definition is that of total factor productivity, which essentially examines the value-added by the factory (an output measure) per unit of input, where the input is defined as a composite of labor, capital, materials, and energy. Few companies track total factory productivity, however, primarily because it does not lend itself very well to managerial interpretation.

Cross-sectional Data. What companies track, almost universally, is direct labor productivity, for historical reasons and for its ready interpretation by managers. Labor productivity, unfortunately, is flawed, partial measure whose value can be augmented by capital-for-labor substitution, among other tactics. Furthermore, concentration on direct labor, a declining fraction of cost for most companies, ignores the increasingly important role played by indirect labor. Indeed, direct labor in most companies constitutes less than 20% of factory output value. Nevertheless, labor productivity measures are widespread and are unlikely to miss, completely, a factory's "true" productivity gains. As such, labor productivity provides the "hardest" quantitative measure that is conveniently procured. Specifically, the labor productivity measure used for this research is the factory's response to the following question:

By whatever measure is commonly used, what has been the approximate average

annual rate of improvement (or decline) of labor productivity for this plant?

Compare last year to the year before. Note that this is a measure of gain from one year to the next rather than a measure of some absolute productivity level (i.e., a measure of a line's slope rather than its position). Absolute productivity levels depend too much on definition and convention and thus cannot easily be used to compare productivity across industries, countries, or even companies. The use of productivity gain eliminates this problem of comparison.

More general measures of productivity, such as those using all employees or those using capital as well as labor inputs, are less biased, but they are less widely used than labor productivity measures and can still problematic. The mail surveys asked plants to supply the productivity gain from any broad measure of productivity that they kept in addition to direct labor productivity. Interestingly, the general measures of productivity gain that they provided are highly correlated with labor productivity gain. Specifically, the general productivity measure used for this research is the factory's response to the following question:

By whatever measure may be used (if any), what has been the approximate average

annual rate of improvement (or decline) of general productivity (more inclusive than

labor productivity) for this plant? Compare last year to the year before.

At the other extreme from these two "hard" number measures of productivity gain are three other, more subjective measures of plant productivity that were asked in the ail surveys. These measures ask the plant to rank itself in three ways:

1. To rank itself in comparison with other company plants

2. To rank itself against other industry plants

3. To rate whether the gain productivity that a plant has achieved has been quickening or

slowing over time (if, in fact, there has been a gain at all).

These are subjective measures, that reveal the perceptions of management about productivity. These perceptions may well color the kinds of productivity programs that factory managers pursue.

Time Series Data. The times series data collected each plant's labor productivity index by quarter. By standardizing each plant's index so that it had a mean of zero and a standard deviation of one, the productivity indexes from each plant could be pooled together without concern for differences in the ways plants scale their productivity measures. By standardizing each of the explanatory variables as well, problems associated with their scaling are likewise avoided.

The detail of the time series data also permitted the development of total factor productivity measures for the five plants. These measures were also standardized with a mean of zero and a standard deviation of one to enable pooling of the time series data.

The analysis of the time series data differs from that of the cross-sectional data. It seeks to explain the standardized levels of productivity in the five plants through time, while the mail survey data analysis examines differences in the productivity gains (or losses) of a broad cross-section of factories. (The distinction is between explaining a line's position and not simply its slope.) The mail survey could not avoid using year-to-year changes in productivity because only a "snapshot" of each plant's productivity performance could realistically be asked for. Dealing with year-to-year gains, however, helps to muffle the "noise" that would more likely abound in, say, quarter-to-quarter comparisons.


Table 2 presents the basic study results for each of the five definitions productivity employed in the mail survey. Each panel of that table summarizes a regression analysis (OLS) using the all-industry, all-location sample (561 total plants). Only statistically significant variables are reported so as to maximize the degrees of freedom for each regression. To conserve space, only estimated signs are reported; full results for all of the tables in this paper, with coefficients, and t-statistics, are available from the author. Consult the appendix for additional comments on the methodology employed. [TABULAR DATA 2 OMITTED]

Perhaps the most interesting results involve the labor productivity gain measure, the study's "hardest number" measure of productivity. As Table 2, Panel A shows, thirteen variables show up as statistically significant at the level of 10% or better.[4] The results tell several stories or themes, all of which make a good deal of intuitive sense:

Process Technology Advance. (1) The lower the average age of the equipment in the plant, the higher the labor productivity gain. (2) The more involved the plant is in pursuing process hardware advances, the greater the productivity gain.

Throughput Time Reduction. The greater the percentage reduction in throughput time (from the release of an order to the factory floor until shipment), the greater the labor productivity gain.

Inventory Reduction. Similarly, the greater the increase in the plant's inventory turns, the greater the labor productivity gain.

Quality and Purchasing. The stronger the functions of quality assurance/control and of purchasing, as perceived by plant management, the greater the labor productivity gain.

Changes in Various Organizational Elements of the Factory. A number of factors loosely tied to the organization of the plant and how productivity ideas are solicited come into play. For example, the more involved the plant is in work force participation, say, by the creation of worker teams, the higher the labor productivity gain. In addition, the fewer the labor grades and the fewer the levels of management of the plant, the higher the labor productivity gain. Interestingly, too, the less that the staff--including industrial engineering, top management, and mid-level management--are involved as primary source of productivity ideas, the higher is the labor productivity gain. (Apparently productivity ideas from other sources are more valuable).

How the Plant Competes. Plants that complete on product performance tend to have higher labor productivity gains than plants that compete on cost, customization, delivery, flexibility, or new product introduction.[5] In addition, plants whose product lines compete in different ways from one another, and are so acknowledged by the plants themselves, also appear to have larger productivity gains.

Utilization of Capacity. A plant which has higher overtime hours per worker also seems to do better on labor productivity. This result is most evident in the Korean subsample (See Schmenner and Rho (1990) for details), but reflects a link between labor productivity and one measure of the capacity utilization at the plant.

Several of these themes carry over to the other panels' regression, particularly to those analyses of the more subjective measures of productivity. For example, the "rank in industry" and the "rank in company" analyses (Table 2, Panels D and E) and the rate of change of productivity analysis (Table, 2 Panel C) are greatly influenced by several of the variables that show up as important in the labor productivity gain results. In particular, throughput time reduction is important in all three of these more subjective measures of productivity, the only variable to be so broadly represented. Two other variables that also capture smoother materials flows--improved layouts, less of the process classes as job shop/batch flow--also turn up as significant, but only in selected regressions.

The extent to which the plant gets involved in hardware advance is important to both the "rank in industry" and "rank in company" results, as well. Rank in industry is also affected by whether the plant patents its technology; if it does, its rank is elevated. The rank of the plant within its own company is elevated if it is able to engage in special deals with equipment suppliers instead of using equipment that is off-the-shelf. Its rank within the company is also enhanced if it is a younger plant. Whether productivity has accelerated or not also seems to be affected by whether the plant has engaged in a lot of the latest computerized hardware advances. The more it has done that, the better off it seems to be.

Several other themes carry over to the subjective measures of productivity. Quality measures (e.g., percent rework) are statistically significant in two of the three subjective measure regressions, with better quality leading to quickened productivity advance and higher perceived rank within the company. Inventory, measured as weeks of inventory held, surfaces as important in the rank within industry results; lower inventories being associated with higher perceived rank within the industry. Some organizational elements such as the importance of productivity ideas from factory floor supervision and customers and the lack of importance of productivity ideas from middle level management also carry over in the results. Efforts at improving "work force morale, effort, and involvement" are important to two of the subjective measures of productivity, while cross-training and broadened job contents are significant in just one of the regressions each.

A number of other factors not seen in the labor productivity regression enter one or more of these subjective measure regressions. The results suggest that productivity suffers when relative strengths lie with inventory control and with materials tracking (attention to the flows themselves seems to be more important), and when programs are instituted to go after materials shortages or overhead directly. Productivity also appears to suffer in selected regressions when product lines age at the factory, when MIS is a relative strength (rather than other functions), and when the plant acts as a feeder to other plants within the company.

When the dependent variable shifts to the general measure of productivity gain (Table 2, Panel B), the character of the results change. General productivity gains seem to be more affected by design criteria than anything else. Variables such as the number of part numbers, a reduced number of engineering change orders, having design capabilities at the plant, and changing the layout of the factory are all important elements in explaining general productivity gains. Also important statistically is the average age of the plant. And, plants with a higher labor cost as a fraction of the sales of the plant also have larger general productivity gains. It seems to be product design and not process design that is effective here because the influence of having your own methods engineering at the plant does not increase productivity but rather seems to decrease it, all other things equal. Moreover, many of these variables are only important to the general productivity gain results, and thus lend a very distinct character to those results. Only one variable, that of technology advance, as represented by the age of the plant's equipment, carries over to others of the regressions run.


The international character of the sample enables us to compare plants across the globe. Table 1 provides such a comparison. The 561 plants were separated into four groups: United States/Canada, Europe, Far East, and Third World. In particular, this meant breaking the IMD international sample apart and sorting its responses. Because the Third World group comprised only 22 plants, it has been omitted from the table. The U.S./Canada group is overwhelmingly U.S., whereas the Far East group is dominated by Korean plants.

The Far Eastern sample has a greater average productivity gain than either the North American or European samples, and its greater gain is statistically significant. Of concern, then, is whether this gain is related to culture, work ethic, governmental programs or regulations, or something else that can be proxied by location, or whether it merely reflects the fact that more Far Eastern factories are pursuing enlightened management practices or have other favorable characteristics (some of which are displayed in Table 1) whose influences can be controlled for. The analysis of Table 3 addresses this question. [TABULAR DATA 3 OMITTED]

The variables used are those found to be statistically significant from Table 2, but with dummy variables introduced that take on the value of 1 if the plant is in a particular location (either Europe for one dummy variable or the Far East for the other) and is zero otherwise. This tests the hypothesis that the intercept of the estimated plane is statistically different for plants from the different continents represented in the sample.

As an examination of Table 3 and the F-tests performed there reveal, however, the estimated dummy variable coefficients are not significantly different from zero at conventional levels. The signs of the estimated coefficients are typically in the expected direction (positive for the Far East and negative for Europe), but their magnitudes are not sufficient to dispel the null hypothesis of zero values for them. What this suggests is that the variables estimated, though accounting for only a fraction of the total plant-to-plant variation, explain enough of the variation so that factory performances, all other things equal, do not differ statistically just because of location. The results suggest that it is plant management and the characteristics of products and processes, not location-related factors, that matter to productivity improvement.


Managers frequently assert that process industry plants (food processing, tobacco, textiles, lumber, pulp and paper, chemicals, petroleum stone/clay/glass, and primary metals) are distinctly different from non-process, assembly-related industries. This data base permits some tests of this assertion.

Table 4 presents this analysis. Column 1 of the table tests the hypothesis that process industries' productivity performance is different by introducing a dummy variable for process industry plants. As can be plainly seen, the estimated dummy variable coefficient is clearly insignificant for each of the five panels of Table 4. A plant's position as a process industry, much as its location, does not appear to affect its labor productivity gain, all other things equal. [TABULAR DATA 4 OMITTED]

A different question is whether the factors affecting productivity gain in process industries are similar to those in non-process industries. One way to examine this issue is to estimate the different models represented in Table 4 with process industry-only data and with non-process industry-only data (Columns 2 and 3 of each panel). The appropriate test for whether the different data can be pooled is the Chow test.(6) The table reports the Chow test for each dependent variable. All tests indicate the appropriateness of pooling process and non-process data. Thus, the policies affecting productivity gain, in general, apply to both types of industry.

Still, a comparison of the different results from the separate regressions is illuminating. For example, consider Table 4, Panel A and the labor productivity measure. Only three variables are statistically significant for both regressions. For both types of industry, (1) throughput time reduction, (2) fewer management levels, and (3) products that compete on performance criteria are associated with higher productivity gains.

On the other hand, there are a number of variables that are statistically significant only for one industry's data or the other. For the process industry plants, the other important variables deal with overtime, quality control/assurance, purchasing, and the development of worker teams. These variables frankly make a lot of sense for process industries where overtime could be expected to be more directly associated with capacity and productivity, where yields and purchased materials are critical to the success of the process, and where many of the early successes with worker teams were seen (Walton (1979)).

The significant variables for the non-process industry sample are two: inventory turns, and recognition that different product lines compete in different ways. These, too, make sense for the more assembly-oriented plants whose processes are apt to have distinct routings and where inventories can build up easily between process steps.

Some sensible, intuitively appealing similarities and differences are also evident in the examination of Table 4, Panel B's results concerning a more general measure of productivity. There, only two variables achieve statistical significance with both data bases: average age of the equipment and number of active part numbers. Both are general influences that one is hard put to tie more to process than to non-process industries. Another two variables--change in engineering change orders (ECOs) and whether new product design is done at the plant--are only significant for the process industry data. Both of these variables reflect the very tight links that exist between product and process in process industries; change the product and the process must be changed as well. As for the non-process industry sample, there were likewise two variables that reached significance: whether methods engineering was done at the plant (if yes, then productivity gain suffered, a counterintuitive result), and labor cost as a percent of value (the higher the labor percent, the greater the general productivity measure gain). Both of these variables are very much ones to be associated with assembly-type operations, where methods and headcount are issues.

When matters turn to the three more subjective measures of productivity, a number of sensible results also follow. Table 4, Panel C investigates whether the plant's productivity has been quickening or slowing over time. There, three variables attain statistical significance with both data samples: percent reduction in throughput time, whether the main product line competes on customization (if so, productivity gain slows), and better materials tracking over the last five years. all of these make sense for both types of process.

As for the variables that matter exclusively to the process industry data, there are six: (1) throughput time (quickening is greater when the throughput time is already great), (2) whether plants directly receive customer orders (if so, productivity gain quickens), (3) perceived relative strength of the planning/scheduling function (the better the function, the quicker the gain), (4) rework percent (the lower the rework, the quicker the productivity gain), (5) recent program to make layout changes (re-layout helps spur gain), and (6) recent program to reduce materials shortages (such a program slows productivity gain). While one could argue why these variables are particularly associated with process industries, the arguments are not strong ones.

One does better in associating the variables whose statistical significance comes only with the non-process industry data. Five variables attain significance: (1) new kinds of (mainly computer-related) equipment, (2) years the major product line was made at the plant (youth quickens gain), (3) perceived relative strength of the inventory control function (strength slows gain), (4) broader job contents, and (5) a program to increase worker morale, effort, and involvement. This group of influences is more plausibly associated with non-process industries.

When it comes to perhaps the two most subjective measures of productivity--rank in company and rank in industry--the mix of variables that are more associated with one type of industry versus another becomes rather chaotic. One is hard pressed to relate convincingly a particular group of statistically significant variables to a particular sample.


Tables 5 and 6 refer to the time series data analysis involving the five auto maker plants. Table 5 deals with the plant's own major measure of productivity, while Table 6 deals with a total factor productivity index computed from data supplied by the plants. An analysis of covariance has been performed for both model specifications, testing the appropriateness of pooling the five time series together.(7) For own measure of productivity, pooling appears to be thoroughly proper [F(23,75) = 0.18 < 1.7, the critical value at 5% significance]. For the total factor productivity measure, however, pooling appears to be less warranted [D(923,75) = 7.43 > 1.7, the critical value]. Nevertheless, both pooled regressions are reported. [TABULAR DATA 5 AND 6 OMITTED]

In general, the results make considerable sense. Recall that the dependent variable is a standardized measure of productivity while the independent variables are also standardized. Consult the appendix for further comments on the methodology. Results shown in Table 5 indicate that productivity increases when:

* the number of shifts declines

* the percent of orders with parts found short decreases

* the percent of factory expedited orders increases

* the throughput time within the factory is reduced

* the replacement value of the plant and its equipment is lower

* the value of inventories rises

* the percent of sales attributable to the main product declines, that is, when there is more

product diversification

* the number of management layers increases

* the number of worker suggestions per worker per year increases

* the scrap rate is reduced

Many of these results make intuitive sense and support the mail survey's cross-section results. Specifically, the results reinforce the favorable impacts of throughput time reduction, quality improvement, and product diversification at the plant.

Some results, however, are puzzling, and could, of course, be an inadvertent result of having only five plants in the data set. For example, why should productivity increases be correlated with increasing levels of inventories or with increasing management layers? Other puzzlers are perhaps explainable. More expediting might help certain measures of productivity. Lower levels of plant and equipment value might indicate the disruptive effect that can initially follow the introduction of new equipment and technology.(8)

Most of the other results, however, make good sense.

Table 6 makes use of the total factor productivity (TFP) index, defined as follows: Total Factor Productivity Index = Value Added by the Factory/Value

of Labor & Capital Inputs Employed

= (Value of Plant Output - Materials Costs - Energy Costs - Outside Services)/(Payroll

+ 0.10 (Value of Plant & Equipment & Inventories))

The labor input is captured by the payroll figure while the 0.10 is an arbitrary rate of return on the services of the stock of capital whose value is captured by the replacement value of the plant, equipment, and inventories.(9) This measure is not particularly correlated with the five factories' own measures of productivity, which is not unusual, given the TFP index's reliance on value added and on all types of inputs.

Using this definition of productivity, Table 6's results indicate that factory productivity is increased when:

* engineering change orders are reduced in number

* sales grow

* new plant and equipment investment levels are lower

* inventories are reduced

* the number of grievances decline [TABULAR DATA 6 OMITTED]

These results, in the main, make intuitive sense. The strongest variable, sales, is a driver of value added and indicates capacity utilization as well, and so can be expected to be heavily correlated with the TFP index measure. These results support lower ECOs and grievances, both readily understandable goals. Unlike the previous results, however, these results support reduced inventories, and that makes much more sense.

Perhaps the only puzzling finding is that higher levels of investment reduce productivity. As mentioned above, such a finding may signal the initial disruptive phase of new technology, which has been remarked by several observers (Hayes and Clark (1986)).


While there are a number of themes that both the cross-sectional and times series data analyses support, there are others for which the analyses provide no support. For example, the traditional shop floor measures of labor efficiency (standard hours earned/actual hours) and of machine utilization have little to do with themes like throughout time, technology advance, or work force involvement.

Data from the Square D Company permit the correlation of labor efficiency with classic measures of labor productivity (output/input) at the factory level. Using data from five Square D factories on labor efficiency and plant-wide productivity over a 30-month time period, two types of analysis were performed:

(1) Mismatches of Labor Efficiency and Productivity. Of the 29 month-to-month changes during the 30-month period, what percent of the time did the labor efficiency and the plant's labor productivity measures move in opposite directions--labor efficiency up (down) and productivity down (up)? If, in fact, improvement in labor efficiency is a good proxy at the factory floor level for labor productivity, then one should expect that there would be relatively few times when one would go up without the other following suit.

(2) Regressions of Labor Efficiency Against Productivity. The monthly data permit the estimation of time series regressions of labor efficiency against plant-wide labor productivity. One should expect a positive and statistically significant relationship between the two.

Table 7 documents the results of these two analyses of the Square D data. As one can see, the degree of mismatches is rather large. Indeed, for one plant, there were more times when the month-to-month movements in the two measures were at odds than there were times when they matched. A similar story can be told with the regressions. Even with just the one independent variable (labor efficiency), one plant's results show a negative correlation, while another shows a statistically insignificant correlation. For no plant is the [R.sup.2] very high for such a time series study.
 Time Period: Monthly data from January 1987 through June 1989
I. Plant Mismatches of Efficiency and Productivity
 Factory #1 31% 9 of 29 month-to-month changes
 Factory #2 38% 11 of 29
 Factory #3 28% 8 of 29
 Factory #4 62% 18 of 29
 Factory #5 45% 13 of 29
II. Estimated Equations: Productivity = a (Efficiency) + constant term
 Labor Efficiency
 Estimate and Constant [R.sup.2]
 t-statistic Term
Factory #1 0.246 127.4 0.08
 (1.60) (11.9)
Factory #2 1.33 -6.64 0.35
 (3.85) (0.23)
Factory #3 0.898 8.77 0.44
 (4.72) (0.63)
Factory #4 -2.06 266.0 0.34
 (3.81) (5.65)
Factory #5 0.370 6.20 0.37
 (4.01) (0.88)
 These results, of course, help to confirm the inappropriateness of labor effic
iency as a proxy
for labor productivity. Factories are better off abandoning such measures in fav
or of others, such
as throughout time, that show more promise of actually fostering productivity ga
 Using consolidated, cross-section survey data from 561 plants from three conti
nents, and
using times series data from five plants of a major domestic auto maker, this pa
per has analyzed
a wide variety of potential influences on factory productivity gain. Some grand
themes have
emerged: technology advance, throughput time reduction, work force involvement,
inventory reduction, and simplified organization all improve productivity. Furt
hermore, using
data from another company (Square D), the use of labor efficiency as a proxy for
has been debunked.
 The breadth of the survey sample (U.S./Canada, Europe, Korea) permitted tests
of two
hypotheses: (1) that location-related factors (e.g., culture, government policie
s) affect factory
productivity, and (2) that status as a "process" industry plant, influences prod
performance. Neither location nor status as a process industry plant, all other
things, equal, is
statistically significant to an explanation of productivity gain.
 The author thanks Professor Boo Ho Rho of Sogang University
in Seoul, Korea for his development of the Korean
data. The author also is grateful to a major domestic auto
maker and the Square D Company for permitting the
publication of the analysis of their plants' data. He is
appreciative as well to the editor and the referees for their very
helpful comments on a previous draft of this paper.
 In empirical research, there are several issues that surface
routinely about the data and methodology employed. The
following discussion is meant to describe further what
was done in the analysis and to address these data and
methodology issues.
 Non-response Bias. The mail survey questionnaires,
organized in the same manner as Table 1, were sent to plant
managers. After pre-testing, there were two waves of U.S.
mailings, one that utilized a listing of Fortune 500 plants
developed from some other research of mine (Schmenner
(1982)), and the other using a random sampling of
manufacturers from across the country. The response rate
in both instances was about 3%, a low but understandable
number given the detail of the survey. A low response
rate naturally brings into question whether the returned surveys
are in some way biased. Two major tests have been made
of the data, both suggesting that there are no obvious biases:
 (1) The U.S. data permitted comparisons of the survey
returns with Census of Manufacturers data and with data
relating to Fortune 500 firms (Schmenner (1982)). The
latter are particularly relevant because the survey purposefully
oversampled large companies. As Table A1 shows, the
relevant chi-square statistics that compare the survey data to the
Census of Manufacturers and to the Fortune 500 data attest
that there are no regional or industry biases of consequence
(see Pindyck and Rubinfeld (1991, p. 520 ff. for a discussion)).
 (2) In addition, the latest arriving 20% of the sample
was compared to the earlier arriving 80%. This was done under
the conventional assumption that if bias exists in the sample
then the latest arriving data should look most like the non-respondent
data. Differences in means t-tests have been
conducted for all of the variables used in the regression. There
were no statistically significant differences in the productivity
measures for the two different groups of data (late versus
earlier arrivals), suggesting that plants complying with
the survey are not particularly better or worse performing than
those plants that did not go to the trouble to fill out the
survey. Only seven of the 53 different variables used in the various
regressions turned up as significantly different at the
5% level, and these seven show no particular trends that would
suggest bias in the sample.
 The IMD surveys were mailed at one time, but because
the mailing was necessarily to IMD participants, many from
the same companies, a true response rate is not known.
The Korean surveys were Korean language translations by
Professor Boo Ho Rho of Sogang University in Seoul
and were mailed out through the auspices of the Korean Traders
Scholarship Foundation. No true response rate is known
for this mailing either. Supplementary census-type data, such as
exist for the U.S., are not available for either the IMD or
Korean data so that tests like those cited in point 1 above are not
possible. Tests of the later arriving data for these two
data bases, like the North American one, reveal no statistically
significant differences attributable to the arrival time and
thus suggest no bias in the returned versus unreturned surveys.
Only six of the 53 variables used were significantly
different in the IMD sample comparison, and only five of the 53 in
the Korean sample comparison. There were no similarities
in the variables so identified either between the IMD and
Korean data bases or between either of them and the U.S. sample.
 Outliers and Residuals. A few outlying values for
productivity gains (e.g., above 100% per year or below -40% per
year) were discarded. Other outliers in the fitted
versus actual dependent variables were examined. All those
observations whose studentized t residuals were in excess
of 1.95 were omitted from runs of the regressions, but no
qualitative differences in the regression results were observed,
suggesting that the remaining outliers do not seem to have
much effect on the estimated results.
 Multicollinearity. With so many potentially significant
variables to investigate, the study was forced to cull many
variables or lose valuable degrees of freedom. Only statistically
significant variables are reported. Variables for analysis
were chosen from those displaying relatively high simple
correlations with the dependent variables and also from those
with strong theoretical or policy-related claims on productivity.
For the regressions reported there are no substantial
multicollinearity problems. The correlations among the
independent variables are low. For the cross-section data (mail
surveys), there are no simpler's in excess of 0.27 and only four
simpler's in excess of 0.20. The correlations for the time
series data are naturally much greater. For the own-measure
regression's independent variables, one correlation (between
part shortages and the percent of sales attributed to the
major product line) reaches 0.65. However, only four others
exceed 0.40. On the other hand, for the total factor
productivity index regression's independent variables, only one
correlation exceeds 0.4.
 Furthermore, the results are robust to minor changes
in the regression specifications run. In developing the results to
report, each of the reported regressions was run with
every other variable entered into it, one at a time, to test both
t-statistics and the robustness of the estimations for
the independent variables already entered. In addition, the
independent variables reported in the paper's tables have
been deleted from each regression, in turn, as another test of the
robustness of the results. In neither case was there any
substantive alteration in the character of the results; estimates and
t-statistics shifted only slightly. The results reported thus
show no evidence of multicollinearity problems.
(1) There is recent, although as yet unpublished, continuing
work in this area at the University of Minnesota and at Ohio
State University.
(2) Time was also examined.
(3) As before, I am deeply grateful of the work of
Professor Boo Ho Rho of Sogang University in Seoul for the
collection of the Korean data. I am indebted also to
the international Institute for Management Development (IMD,
formerly IMEDE) in Lausanne, Switzerland, for
sponsoring the gathering of the mainly European data.
(4) A significance level of 10%, more generous than the
conventional 5%, was used for this explanatory work so that the
chance of committing a Type II error was reduced.
(5) Competition on "quality" was not included as a
response because of both the vagueness of the term to many and
because it was thought that nearly all would acknowledge it as important.
(6) Consult Pindyck and Rubinfeld (1991) p. 115 ff.
(7) Consult Pindyck and Rubinfeld (1991) p. 225.
(8) The effects of lagging the investment variable were
explored, under the supposition that this result might be explained
by lagged effects that only provide benefits after
some time has passed. To test this, the investment variable was
lagged a quarter and then two quarters, and both
variables introduced to the model. F-tests confirm their influence as
a group in both regressions. For the own measure
model, only the lagged one quarter variable achieves statistical
significance. However, it carries a negative sign, while
the other two carry positive signs. The simple correlations of
all three investment variables with the own
productivity measure are negative, however, suggesting that the sign
reversals are likely caused by the multicollinearity
inherent in the lagged versions of the investment variables.
In the total factor productivity measure model,
only the non-lagged investment variable is significant, and
negative. The one quarter lagged variable is positive,
while the two quarter lagged variable is negative. Again, the
simple correlations are all negative, suggesting
that multicollinearity is at play with the estimated signs.
(9) Sometimes, however, only book values of the capital
assets of the plant were available, and this can color the results.
Hayes, Robert H., and Kim B. Clark. "Exploring the Sources
of Productivity Differences at the Factory Level." In The
Uneasy Alliance: Managing the Productivity-Technology
Dilemma, K. B. Clark, R.H. Hayes, and C. Lorenz (eds.)
Boston: HBS Press, 1985, 151-194.
Hayes, Robert H., and Kim B. Clark. "Why Some Factories
Are More Productive Than Others." Harvard Business
Review, vol. 64, no. 5, September-October 1986, 66-73.
Hayes, Robert H., and Kim B. Clark. "Explaining Observed
Productivity Differentials between Plants: Implications for
Operations Research." Interfaces, vol. 15, no. 6,
November/December 1985, 3-14.
Kleinbaum, David G., Lawrence L. Kupper, and Keith E.
Muller. Applied Regression Analysis and Other Multivariable
Methods, 2nd ed., Boston: PWS-Kent Publishing Company, 1988.
Pindyck, Robert S., and Daniel L. Rubinfeld. Econometric
Models and Economic Forecasts, 3rd ed. New York:
McGraw-Hill, Inc., 1991.
Schmenner, Roger W., and Randall L. Cook. "Explaining
Productivity Differences in North Carolina Factories."
Journal of Operations Management, vol. 5, no. 3, May
1985, 273-89.
Schmenner, Roger W. "Comparative Factory Productivity."
Report to the U.S. Department of Commerce, Economic
Development Administration, Research and Evaluation
Division, July 1986.
Schmenner, Roger W., "An International Examination of
Factory Productivity." Report to the Institute for Management
Development, Lausanne, Switzerland, June 1987.
Schmenner, Roger W., "The Merit of Making Things Fast."
Sloan Management Review, vol. 30, no. 1, Fall 1988,
Schmenner, Roger W. "Behind Labor Productivity Gains in
the Factory." Journal of Manufacturing and Operations
Management, vol. 1, no. 4, Winter 1988, 323-338.
Schmenner, Roger W., and Rho, Boo Ho. "An International
Comparison of Factory Productivity." International Journal
of Operations and Production Management, vol. 10, no. 4,
1990, 16-31.
Schmenner, Roger W. Making Business Location Decisions.
Englewood Cliffs. NJ: Prentice-Hall, Inc., 1982.
Walton, Richard E. "Work Innovations in the United States."
Harvard Business Review, vol. 57, no. 4, July/August
1979, 88-98.
 Roger W. Schmenner
Indiana University School of Business
COPYRIGHT 1991 American Production and Inventory Control Society Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1991 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Schmenner, Roger W.
Publication:Journal of Operations Management
Date:Apr 1, 1991
Previous Article:Human resource issues in cellular manufacturing: a sociotechnical analysis.
Next Article:The capacitated dynamic lot-sizing problem with startup and reservation costs: a forward algorithm solution.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters