Comparing earnings inequality using two major surveys.Some previous research suggests that discrepancies exist between the National Longitudinal lon·gi·tu·di·nal adj. Running in the direction of the long axis of the body or any of its parts. Survey of Youth and the Current Population Survey in terms of earnings trends; when the sample is limited to full-time full-time adj. Employed for or involving a standard number of hours of working time: a full-time administrative assistant. full , year-round workers, however, the discrepancies are largely eliminated Much of the research on the growing dispersion dispersion, in chemistry dispersion, in chemistry, mixture in which fine particles of one substance are scattered throughout another substance. A dispersion is classed as a suspension, colloid, or solution. of earnings has relied on the March supplement to the Current Population Survey (CPS (1) (Characters Per Second) The measurement of the speed of a serial printer or the speed of a data transfer between hardware devices or over a communications channel. CPS is equivalent to bytes per second. ). As the research questions have turned to such issues as job instability instability /in·sta·bil·i·ty/ (-stah-bil´i-te) lack of steadiness or stability. detrusor instability and long-term Long-term Three or more years. In the context of accounting, more than 1 year. long-term 1. Of or relating to a gain or loss in the value of a security that has been held over a specific length of time. Compare short-term. wage growth, however, the focus often has shifted to longitudinal surveys, such as the Panel Study of Income Dynamics (PSID PSID Panel Study of Income Dynamics PSID Panel Study on Income Dynamics PSID Pounds per Square Inch Differential PSID Photon Stimulated Ion Desorption PSID Product Support Integration Directorate PSID Private System Identification )(1) and the National Longitudinal Surveys The National Longitudinal Surveys (NLS) are a set of surveys conducted by the US Department of Labor's Bureau of Labor Statistics, designed to gather information at multiple points in time on significant life events of several population samples of US citizens, especially their (NLS NLS - Native Language System ).(2) In a recent unpublished but widely cited paper,(3) Peter Gottschalk Gottschalk or Gottschalck (both: gôt`shälk), d. c.868, German theologian; son of the count of Saxony. He was placed as a boy in the monastery of Fulda (c.822). and Robert Robert, Henry Martyn 1837-1923. American army engineer and parliamentary authority. He designed the defenses for Washington, D.C., during the Civil War and later wrote Robert's Rules of Order (1876). Noun 1. A. Moffitt Moffitt is a reference to Clan Moffat, on of the oldest Scottish clans. It may also refer to: Cancer Research
1. in epidemiology, a group of individuals sharing a common characteristic and observed over time in the group. 2. show roughly the same trends as the CPS, although the magnitudes are quite different. For the later NLS cohort, however, known as the National Longitudinal Survey of Youth 1979 (NLSY NLSY National Longitudinal Survey of Youth (USA) 79), Gottschalk and Moffitt find both significantly lower variance The discrepancy between what a party to a lawsuit alleges will be proved in pleadings and what the party actually proves at trial. In Zoning law, an official permit to use property in a manner that departs from the way in which other property in the same locality in reported annual earnings and a negative trend in variance over time (1979-1988)--at least for high school graduates. In addition, a more recently published paper using different methodology finds a similar discrepancy DISCREPANCY. A difference between one thing and another, between one writing and another; a variance. (q.v.) 2. Discrepancies are material and immaterial. .(5) Because the findings of these studies stand in sharp contrast to the well-known well-known adj. 1. Widely known; familiar or famous: a well-known performer. 2. Fully known: well-known facts. "stylized styl·ize tr.v. styl·ized, styl·iz·ing, styl·iz·es 1. To restrict or make conform to a particular style. 2. To represent conventionally; conventionalize. fact" that the variance in earnings was increasing substantially during the 1980s, serious questions may be raised about the validity of the NLSY79 for research on the topic of recent trends in earnings inequality inequality, in mathematics, statement that a mathematical expression is less than or greater than some other expression; an inequality is not as specific as an equation, but it does contain information about the expressions involved. . This article focuses on the comparison between the NLSY79 and the CPS, updating the Gottschalk-Moffitt analysis to 1994, the final year of data collection for the NLSY79 cohort. Because Gottschalk and Moffitt report few discrepancies in the trends for high school dropouts, the analysis is restricted to high school graduates. The article begins by replicating the Gottschalk-Moffitt analysis in order to verify (1) To prove the correctness of data. (2) In data entry operations, to compare the keystrokes of a second operator with the data entered by the first operator to ensure that the data were typed in accurately. See validate. the discrepancies in reported earnings between the two sets of data. Next, exploratory data analysis Exploratory Data Analysis - (EDA) [J.W.Tukey, "Exploratory Data Analysis", 1977, Addisson Wesley]. and respecified regression regression, in psychology: see defense mechanism. regression In statistics, a process for determining a line or curve that best represents the general trend of a data set. models are used to compare the trends and patterns, and to look for potential sources of the discrepancies. The final section discusses the implications of the findings for the validity of the two samples. Data and methods The present study generally follows the conventions adopted by Gottschalk and Moffitt. For their benchmark analyses, they select white males in the civilian noninstitutionalized adj. 1. not committed to an institution; - op people. Opposite of institutionalized nt>. Adj. 1. noninstitutionalized - not committed to an institution noninstitutionalised population and divide the samples into cells defined by single years of age (from 16 to 31 years), level of education (less than a high school education, high school graduate or more), and survey year (1979-88).(6) Nominal annual earnings are adjusted for inflation and are expressed in constant (1982) dollars. Also, to avoid topcoding issues and reduce the problem of earnings nominally falling below minimum wage, the top and bottom 5 percent of the values are trimmed out within each cell. Because the trimming is based on the percentiles within cells rather than across the entire sample, the cells are the unit of analysis. As in the earlier paper, for the regression analyses, the CPS and NLSY79 samples are restricted to respondents In the context of marketing research, a representative sample drawn from a larger population of people from whom information is collected and used to develop or confirm marketing strategy. who were aged 20 years or older in the survey year and whose earnings and number of weeks worked during the previous calendar year both were positive. The dependent variable is the within-cell standard deviation In statistics, the average amount a number varies from the average number in a series of numbers. (statistics) standard deviation - (SD) A measure of the range of values in a set of numbers. of trimmed real log annual earnings in the year prior to the interview. Updating the Gottschalk-Moffitt analysis beyond 1988 requires some changes to the sample selection criteria due to changes in survey coding procedures that have taken place since then. In addition, to focus the sample more tightly on a homogeneous The same. Contrast with heterogeneous. homogeneous - (Or "homogenous") Of uniform nature, similar in kind. 1. In the context of distributed systems, middleware makes heterogeneous systems appear as a homogeneous entity. For example see: interoperable network. set of white males, some new exclusions are adopted. The following tabulation tab·u·late tr.v. tab·u·lat·ed, tab·u·lat·ing, tab·u·lates 1. To arrange in tabular form; condense and list. 2. To cut or form with a plane surface. adj. Having a plane surface. compares the sample selection criteria used in the present analysis with those used by Gottschalk and Moffitt in their study.
Criteria Gottschalk-Moffitt Updated analysis
Years 1979--1988 1979-1994
Age range 16-21 in 1979 16-21 in 1979
Race White White, non-Hispanic
Enrollment Employment status No student
recode-based exclusion exclusion
Earnings Positive Positive
Regression sample:
Age 20 years and older 20 years and older
Weeks worked Positive Positive
The most important difference in the criteria used here concerns the exclusion of students. On the basis of the "employment status recode Verb 1. recode - put into a different code; rearrange mentally; "People recode and restructure information in order to remember it" rearrange - put into a new order or arrangement; "Please rearrange these files"; "rearrange the furniture in my room" " variable, Gottschalk and Moffitt exclude CPS and NLSY79 respondents who reported school attendance as their major activity during the survey week. But the coding for this variable in the CPS was changed in 1988 and it no longer identifies school attendance as a unique status. To preserve consistency across the time series, therefore, this analysis does not directly exclude students in this way. The overall impact of the change is relatively small, though, because several of the other exclusions (positive earnings and number of weeks worked, for example) capture much of the same population.(7) For each data set, descriptive regression analyses similar to those used in the earlier study were conducted to compare the trends in earnings across the different samples. Let [y.sub.at], be the standard deviation of the log annual wages for workers age a in year t. The model fit by Gottschalk and Moffitt is a simple linear specification: [A] [y.sub.at] = [[Beta].sub.0] + [[Beta].sub.1]a + [[Beta].sub.2]t + [[Epsilon 1. (language) EPSILON - A macro language with high level features including strings and lists, developed by A.P. Ershov at Novosibirsk in 1967. EPSILON was used to implement ALGOL 68 on the M-220. ].sub.at] a = 20, ..., 36; t = 79, ..., 94 where [[Beta].sub.1] and [[Beta].sub.2] are the coefficients for the linear effects of age and year, respectively. The present analysis extends the earlier study in two ways. First, the regression model is respecified and two alternative specifications are examined: a nonparametric nonparametric said of statistical techniques which do not depend on the data having a normal or some other definable distribution. model for the age term and a random-effects model to capture the longitudinal sample dependence in the NLSY79. The regression residuals for model A show a marked curvilinear curvilinear a line appearing as a curve; nonlinear. curvilinear regression see curvilinear regression. pattern in age that is roughly parabolic par·a·bol·ic also par·a·bol·i·cal adj. 1. Of or similar to a parable. 2. Of or having the form of a parabola or paraboloid. in nature. The time trend is of primary interest here, rather than the effects of age. Given the correlation between year and age in these samples, however, the age effect must be specified properly to obtain an accurate estimate of the time trend. As the linear age specification compromises the interpretation and statistical significance of the coefficients of both linear coefficients, the model is respecified using a nonparametric age effect, as follows: [B] [y.sub.at] = [[Alpha].sub.e] + [[Beta].sub.a] + [Beta]t + [[Epsilon].sub.at] a = 20,..., 36; t = 79,....,94 where [[Beta].sub.0], ..., [[Beta].sub.36] are coefficients for each age and B is the regression parameter (1) Any value passed to a program by the user or by another program in order to customize the program for a particular purpose. A parameter may be anything; for example, a file name, a coordinate, a range of values, a money amount or a code of some kind. for the linear time trend. It is important to note that the two previous studies have treated both the CPS and the NLSY79 as cross-sectional cross section also cross-sec·tion n. 1. a. A section formed by a plane cutting through an object, usually at right angles to an axis. b. A piece so cut or a graphic representation of such a piece. 2. surveys, although the latter is a longitudinal survey. There are eight cohorts in the NLSY79, defined by respondent's age in 1979, and each cohort is followed across the entire 16 years of the series. Observations from the same cohort in the NLSY79 are likely to be correlated cor·re·late v. cor·re·lat·ed, cor·re·lat·ing, cor·re·lates v.tr. 1. To put or bring into causal, complementary, parallel, or reciprocal relation. 2. across time, a fact not taken into account in the Gottschalk-Moffitt analysis, the study by Thomas (language) Thomas - A language compatible with the language Dylan(TM). Thomas is NOT Dylan(TM). The first public release of a translator to Scheme by Matt Birkholz, Jim Miller, and Ron Weiss, written at Digital Equipment Corporation's Cambridge Research Laboratory runs MaCurdy and others (cited earlier), or in the models (A and B) shown above. The cohort sample dependence can be modeled in one of two ways--as a fixed effect or as a random effect. Adding a fixed effect to either model A or model B is not possible because the parameters for age, year, and cohort are perfectly confounded (cohort -- year minus age). A random-effect specification is therefore required and also is more appropriate from a substantive standpoint The Standpoint is a newspaper published in the British Virgin Islands. It was originally published under the name Pennysaver, largely as a shopping-coupon promotional newspaper, but since emerged as one of the most influential sources of journalism in the . The interest here is not in the cohort effects The term cohort effect is used in social science to describe variations in the characteristics of an area of study (such as the incidence of a characteristic or the age at onset) over time among individuals who are defined by some shared temporal experience or common life as indicators of inherent differences among specific age-year groups. The cohorts are simply samples from their populations, and this study seeks to capture the covariance Covariance A measure of the degree to which returns on two risky assets move in tandem. A positive covariance means that asset returns move together. A negative covariance means returns vary inversely. in these samples over time, rather than an estimate of a cohort-specific level effect. Therefore, model B is respecified for the NLSY79 to include a random effect for cohort, as follows: [C] [y.sub.atc] = [Alpha] + [[Beta].sub.a] + [[Beta]t + [[Epsilon].sub.ic] a = 20,...,36; t = 79,...,94; c = 1,...,8; [[Epsilon].sub.ic] [[Phi].sub.c] + [[Sigma SIGMA - A scientific visual programming environment from NASA. http://fi-www.arc.nasa.gov/fia/projects/sigma/. ].sub.ic] where [B.sub.20], ..., [[Beta]t.sub.36] are coefficients for each age, [Beta] is the coefficient coefficient /co·ef·fi·cient/ (ko?ah-fish´int) 1. an expression of the change or effect produced by variation in certain factors, or of the ratio between two different quantities. 2. for the linear effect of year, and [[Phi].sub.1], ..., [[Phi].sub.8] are random variance components for each cohort. Because it requires no assumptions about the parametric See parametric modeling, parametric symbol and PTC. form of the random cohort effects, a generalized gen·er·al·ized adj. 1. Involving an entire organ, as when an epileptic seizure involves all parts of the brain. 2. Not specifically adapted to a particular environment or function; not specialized. 3. estimating equation (GEE gee 1 n. The letter g. gee 2 interj. Used to command a horse or ox to turn to the right. intr.v. ) is used to fit the model.(8) For all of the linear models, weights are used to reflect the differing variances of the [y.sub.at] component of the model.(9) In the GEE models, the variance-covariance weight matrix includes covariance estimates in the off-diagonal cells to adjust for the longitudinal cohort sample dependence. All models are fit using the S.PLUS statistical program.(10) The second way in which the present study extends the Gottschalk-Moffitt analysis is by reexamining the discrepancies in earnings dispersion by labor force status. Gottschalk and Moffitt use several indicators as proxies of labor force attachment in an attempt to explain the discrepancy in earnings trends: the employment status recode variable, more than 40 weeks worked in the past year, and age 23 years and older (presumably pre·sum·a·ble adj. That can be presumed or taken for granted; reasonable as a supposition: presumable causes of the disaster. to exclude most college-age students). The present study takes a more direct approach, subdividing the sample into two groups: full-time, year-round workers (FTFY FTFY Fixed That for You ) and others (non-FTFY). The FTFY group comprises those who worked 35 or more hours per week and 50 or more weeks per year during the previous calendar year; the non-FTFY group comprises those who had positive earnings and hours worked but who did not work work full time and year round. For the CPS, the constructed variable that identifies this status is used, and for the NLSY79, hours and weeks are selected directly. The definition is the same in both samples. The idea here, as in the earlier study, is to compare workers with relatively strong attachments to the labor force with workers who are less attached to the labor force. Results Tables 1 and 2 provide summary statistics for labor force attachment and annual earnings for workers in both data sets in 1979, the first year of the series. The sample selections reflect the updated analysis criteria and can be compared with the corresponding tables in the paper by Gottschalk and Moffitt. Table 2 shows patterns similar to those found in the earlier study--a significantly larger portion of the NLSY79 sample reports working 40 weeks or more per year. While fairly pronounced in 1979, this discrepancy in the number of weeks worked during the year declines in subsequent years.
Table 1. Basic descriptive statistics for 1979 survey year
High school
graduates Unweighted N
(in percent)
Age
NLSY79 CPS NLSY79 CPS
Total (all ages) 44.7 57.8 796 3,261
16 0.4 0.2 1 4
17 .9 .5 4 30
18 45.6 47.5 145 507
19 79.3 80.3 218 903
20 86.7 88.4 224 885
21 87.4 87.5 204 932
Among high school graduates
Percent working Percent working
at least 1 week 40 or more weeks
Age during the year during the year
NLSY79 CPS NLSY79 CPS
Total (all ages) 95.6 92.5 52.9 48.8
16 ... ... ... ...
17 19.0 ... ... ...
18 96.3 90.3 39.2 37.3
19 95.2 91.7 49.0 43.4
20 94.4 93.5 62.3 50.1
21 96.6 93.9 54.6 59.1
Percent working
full time, year round
Age
NLSY79 CPS
Total (all ages) 26.1 28.4
16 ... ...
17 ... 3.4
18 10.5 15.3
19 23.5 22.6
20 36.4 32.4
21 27.2 37.1
Table 2. Basic Income statistics for survey year 1979
Unweighted N Income
mean
Age
NLSY79 CPS NLSY79 CPS
All workers
16 1 2 1,221
17 2 28 2,608 3,071
18 118 601 3,814 4,163
19 198 1,100 6,120 5,819
20 214 1,160 8,373 6,643
21 202 1,230 8,812 8,991
Full-time, year-
round workers
16 0 0 ... ...
17 0 1 ... 3,497
18 12 93 5,380 7,547
19 45 245 10,067 10,414
20 83 385 11,413 10,823
21 51 481 13,648 13,374
Covariance Log income Standard
mean deviation
Age
NLSY79 CPS NLSY79 CPS NLSY79 CPS
All workers
16 ... ... ... 7.11 ... 0.00
17 463 3,323 7.84 7.54 .35 1.08
18 1,416 1,621 8.02 8.12 .73 .69
19 2,716 2,817 8.45 8.39 .80 .80
20 2,661 2,938 8.84 8.53 .67 .80
21 3,793 3,768 8.82 8.83 .79 .81
Full-time, year-
round workers
16 ... ... ... ... ... ...
17 ... ... ... ... ... ...
18 839 805 8.48 8.87 .55 .36
19 1,354 1,012 9.13 9.18 .47 .42
20 1,607 1,086 9.24 9.23 .51 .38
21 1,466 1,581 9.46 9.42 .39 .48
NOTE: Statistics are calculated using sample weights and 5 percent trim of top and bottom earnings. Unweighted N reflects post-trim cell values. Despite the difference in reported number of weeks worked, the earnings figures in table 2 are quite similar across the two samples. There are no systematic differences in either means or variances. The numerical numerical expressed in numbers, i.e. Arabic numerals of 0 to 9 inclusive. numerical nomenclature a numerical code is used to indicate the words, or other alphabetical signals, intended. values are different than those reported by Gottschalk and Moffitt, due largely to the inclusion here of students who had been excluded in the earlier study on the basis of the employment status recode variable. The bottom portion of the table shows the statistics for FTFY respondents--a group likely to exclude such students--and here the two samples become very close. The trends in earnings variances over time for the two samples are shown in chart 1. They show a general decrease in earnings dispersion with age, and this pattern is much stronger than the trend over time within specific age groups. The NLSY79 estimates are more variable, reflecting the smaller sample sizes. Net of the differences in variability between the two samples, the greatest differences between them occur within the younger age groups--those aged 19 to 24 years. These differences are not very systematic, and in particular, they do not appear to take the form of consistently stronger increasing trends over time in the CPS. There is some convergence between the two samples for the older respondents, but the earnings dispersion for the NLSY79 is about 10 percent lower, on average, than for the CPS. By contrast, the cell median incomes in the NLSY79 are consistently about 20 percent higher than the corresponding CPS cell means (data not shown here). Once the two samples of respondents settle into their prime working years, then, the annual earnings reported in the NLSY79 are both higher and less variable than those reported in the cps. [Chart 1 OMITTED] The standard deviations are modeled by reverting re·vert intr.v. re·vert·ed, re·vert·ing, re·verts 1. To return to a former condition, practice, subject, or belief. 2. Law To return to the former owner or to the former owner's heirs. to cells defined by survey year and single year of age. Much like the Gottschalk--Moffitt study, attention here is restricted to those aged 20 years and older, with positive weeks worked in the previous calendar year. The results are displayed in table 3. All coefficients are multiplied mul·ti·ply 1 v. mul·ti·plied, mul·ti·ply·ing, mul·ti·plies v.tr. 1. To increase the amount, number, or degree of. 2. Mathematics To perform multiplication on. by 10 to be consistent with the values reported by Gottschalk and Moffitt. The coefficients can be interpreted as the change in standard deviation over a 10-year period.
Table 3. Regression results
Sample restriction and model CPS NLSY79
Gottschalk-Moffitt analysis:
CPS--not in school 0.019 -0.038
NLSY79--
nonenrolled ... .038
NLSY79--
23 years and older ... -.100
Updated analysis:
1979-88 only
A .150 -.124
B .020 -.093
C ... -.089
All workers, 1979-94
A -.049 -.165
B .009 -.085
C ... -.092
Full-time year-round workers, 1979-94
A .025 -.030
B .032 -.020
C ... .036
Part-time, part-year workers, 1979-94
A .030 -.126
B .042 -.096
C ... -.116
Full-time, year-round workers, 1979-94,
excluding self-employed
A .033 -.019
B .041 -.004
C ... .027
NOTE: Model A specifies linear effects for both age and ,ear, model B specifies a non-parametric age effect, and Model C includes a random effect for longitudinal cohort dependence in the NLSY79. The results obtained by Gottschalk and Moffitt are shown in the first three rows of the table for comparison. Consider first their results based on the employment status recode schooling exclusion. For the cps, they find a positive but not significant upward trend in earnings dispersion, while the corresponding trend for the NLSY79 is negative and also not significant. Using a more specific measure of school enrollment over the past year that is available in the NLSY79 to exclude students in that sample, they find the coefficient for the trend in dispersion changes sign and becomes as strongly positive as it had been negative, though still not significant. Further restricting this NLSY79 sample to those aged 23 years and older, they find the coefficient changes sign again and is now much more strongly negative than it had been, though still not significant. The Gottschalk-Moffitt estimate of the time trend is thus extremely sensitive to the sample exclusions. The same is true in the present analysis, in part due to the relatively small number of observations in each cell after the screens for positive earnings and weeks worked and the 10-percent trimming. This makes for a high level of instability in the cell-specific estimates of the earnings variance, and these in turn have a large impact on the within-age trend estimates. The latter is due to the interaction between the model, which estimates the time trend within age, and the structure of the sample. While the two surveys cover 16 years, age groups are observed for, at most, 8 years, and the average for persons aged 20 years and older is 6.3 years. The moving cohort window is thus not an ideal structure for capturing trends within age over time. When drawing inferences about the discrepancies between the two samples, it should be kept in mind that the estimates are not particularly robust. The remaining rows in table 3 present the results from the updated analysis. In the first set, we restrict the sample to the years used by Gottschalk and Moffitt, 1979-1988. The differences between the results for model A and the results in the first row of the Gottschalk-Moffitt figures reflect the difference in the sample restrictions between the two analyses--namely, the inclusion in this analysis of students who were excluded from the earlier study on the basis of the CPS employment status recode, as well as the exclusion here of Hispanics. The impacts are not dramatic, with the CPS coefficient becoming slightly less positive under the new sample restrictions. The NLSY79 coefficient becomes more negative and now also is statistically significant, though in magnitude it still lies within the range of estimates reported in the earlier study. When a nonparametric specification for age is adopted in model B, the discrepancy declines--the CPS coefficient increases modestly, and the NLSY79 coefficient becomes much less negative. When the random effect for the longitudinal cohort dependence in the NLSY79 (model C) is added, the coefficient for the time trend again becomes slightly less negative, and now it is about 30 percent lower than the initial estimate in model A. While the numerical results obtained in the earlier study are not replicated exactly, the general pattern is replicated, showing an increasing trend for earnings dispersion in the CPS and a decreasing trend for the NLSY79. The magnitude of the discrepancy and of the negative trend in the NLSY79 becomes smaller in both of the respecified models. The next set of results shown in table 3 (labeled all workers) updates the analysis to 1994. For the cPS, the trend in earnings dispersion is now significantly negative in model A, as is the trend for the NLSY79. With the nonparametric age effect, the sign of the CPS coefficient changes to become positive (although weakly weak·ly adj. weak·li·er, weak·li·est Delicate in constitution; frail or sickly. adv. 1. With little physical strength or force. 2. With little strength of character. so and not significant), while the magnitude of the NLSY79 coefficient is still negative but reduced by about half. Adding the random effect to the NLSY79 slightly increases the magnitude of the negative trend, but it is still 40 percent lower than the estimate under the initial model. Respecifying the model once again reduced the discrepancy between the two samples. The results from model C are graphically displayed in chart 2. The top panel plots the nonparametric age-effect estimates. The results show that earnings dispersion is highest among the young, and it falls steeply through the mid-twenties age groups. For the cps, dispersion then begins to rise slightly, while for the NLSY79, the decline continues through the early-thirties age groups, though less steeply, and then also begins to rise. The nonlinearity for the CPS is more pronounced, which helps to explain why the nonparametric specification in Model B has a relatively larger impact on the trend coefficient for that sample. [Chart 2 OMITTED] The bottom panel of chart 2 shows the partial regression plot of earnings dispersion by year after adjusting for age. The trend lines are nonparametric local-linear estimates. As can be seen, the CPS trend is modestly positive. The plot for the NLSY79, by contrast, clearly shows a negative trend. Note, however, the large residual variation. The magnitudes of the time trends for both samples are modest relative to the residual variability. Next, the analysis is restricted to full-time, year-round workers in order to determine whether the discrepancies in earnings dispersion between the two samples persist among the core group of workers with the strongest attachment to the labor force. This group becomes an increasingly larger share of the two samples over time, rising from about 35 percent of the regression-eligible sample in 1979 to 80 percent in 1994. If the trend differential persists for these workers, then it is a fundamental and pervasive pervasive, adj indicates that a condition permeates the entire development of the individual. discrepancy. If not, then the samples are comparable for the core workers, and some progress has been made in narrowing down the possible sources of the problem. The trend coefficient under model A reproduces the discrepancy observed above, but the negative trend for the NLSY79 is substantially smaller than in all of the previous analyses. The estimates from model B are consistent with the earlier pattern--that is, the discrepancy narrows as the trend becomes more positive for the CPS and less negative for the NLSY79. When the random effect for the sample dependence in model C is added, however, the NLSY79 coefficient changes sign, becoming strongly positive and similar in magnitude to the cps coefficient, though not statistically significant. Under model C, then, both samples of full-time, year-round workers show a positive trend in earnings dispersion of comparable magnitude. The results for the other (non-FTFY) workers show the opposite pattern, with the discrepancy very large under model A and virtually unchanged under model C. For these workers, opposite trends are seen in earnings dispersion for the two samples--dispersion grows over time in the cps, while it declines over time in the NLSY79. The pattern of statistical significance is also different for this subgroup sub·group n. 1. A distinct group within a group; a subdivision of a group. 2. A subordinate group. 3. Mathematics A group that is a subset of a group. tr.v. , with the NLSY79 trends testing highly significant and the CPS trends testing only modestly significant. The age effects and partial regression plots for model C for the full-time, year-round workers and for the other workers are shown in chart 3. The pattern of higher dispersion for older NLSY79 respondents also is visible here in both subgroups. The smoothed trend lines are clearly different, however, with the FTFY workers in both the CPS and LSY LSY Lismore, New South Wales, Australia - Lismore (Airport Code) 79 samples now showing a weak positive trend. The residual variability also differs: it is now lower for the FTFY workers and higher for the non-FTFY workers. The smoothed trend lines do not tell an entirely unambiguous story--when the endpoints are excluded, a different trend sometimes emerges. The regression line Noun 1. regression line - a smooth curve fitted to the set of paired data in regression analysis; for linear regression the curve is a straight line regression curve would be even more strongly influenced by the high leverage points at the extremes, simply reinforcing the earlier point that caution is appropriate when drawing inferences from any of the trend coefficients estimated from these samples. [Chart 3 OMITTED] One final analysis was conducted in which the self-employed self-em·ployed adj. Earning one's livelihood directly from one's own trade or business rather than as an employee of another. self were excluded. This is a group known to have highly variable earnings. They are almost universally excluded in studies of earnings inequality because their earnings determination process is fundamentally different from that of wage and salary workers. Excluding the self-employed, the pattern obtained is basically the same as that of the full sample of FTFY workers: in the final specification of model C, both samples again show a positive trend of similar magnitude in earnings dispersion over time. These analyses suggest that the earnings dispersion discrepancy found by Gottschalk and Moffitt results largely from the specification of their regression model as well as a trend that appears to be driven by workers who do not work full time and year round. To examine the latter, chart 4 shows the trends in earnings dispersion by age-year cell separately for and non-FTFY workers.(11) The trends for FTFY workers look similar for the two samples--that is, both groups show a modest upward trend. The age effects discussed earlier (see chart 1) are completely absent here. In the graph for non-FTFY workers, by contrast, the CPS shows a fairly stable pattern of earnings dispersion over time, while the trend for the NLSY79 is somewhat negative. This clearly is what is driving the negative trend in the NLSY79 data when both groups of workers are combined. For non-FTFY workers, the age differences are absent as well. Thus, what at first appears to be an age effect in the graph for all workers actually is a composition effect--as age increases, the majority of workers shift from non-FTFY status to working full time and year round. [Chart 4 OMITTED] To better understand the nature of these discrepancies, it is useful to look at estimates of the distributions themselves. Chart 5 shows the 1979 earnings densities for the two samples as an example.(12) The top panel corresponds to all workers. While the two distributions are similar at the higher earnings levels, the CPS sample has a longer, denser lower tail than the NLSY79 sample. The bottom panel shows the corresponding distributions for non-FTFY workers. The CPS distribution is strongly downshifted, indicating lower levels of reported earnings compared with the NLSY79, and the bottom tail of the distribution for these workers reaches much further down the earnings scale. The location of the lower tail of the non-FTFY earnings density, from about 6 to 8 on the log scale, corresponds exactly to the location of the lower tail differences in the distribution for all workers. The plot for FTFY workers, not shown here, looks much like the plot for all workers, without the greater relative density in the lower tail of the CPS. [Chart 5 OMITTED] This lower tail discrepancy becomes more pronounced over time, as can be seen by the 90:50 and 50:10 earnings ratios for non-FTFY workers shown in chart 6. The 50:10 ratio for the two samples is relatively similar at the start of the series, but the CPS ratio increases over time while the NLSY79 ratio declines. Given the consistently lower median reported earnings in the CPS, the rise in the 50:10 ratio implies an increasingly longer tail at the bottom of the distribution than that observed in the NLSY79. The 90:50 ratios are more similar for the two samples, with both showing a downward trend over time, though the timing of the decline is different. The variance differential between the two samples is thus being driven primarily by the discrepancies in the lower tails. Specifically, it is being driven by the longer lower tail of the CPS non-FTFY earnings distribution. [Chart 6 OMITTED] Discussion The discrepant dis·crep·ant adj. Marked by discrepancy; disagreeing. [Middle English discrepaunt, from Latin discrep findings in the trends in annual earnings dispersion between the CPS and the NLSY79 appear to be a function of the model specification and the non-Fret workers. Regression diagnostics (1) Software routines that test hardware components (memory, keyboard, disks, etc.). Diagnostics are often stored in ROM chips and activated on startup. (2) Error messages in a programmer's source code that refer to statements or syntax that the compiler or assembler clearly show that a linear specification for age is not appropriate, and fitting a nonparametric effect reduces the discrepancy in the estimated dispersion trends by one-third to one-half. Treating the two samples as cross-sectional, thus ignoring the longitudinal cohort dependence in the NLSY79, also is not appropriate. Modeling the cohort dependence in the NLSY79 changes the estimates of the dispersion trend, especially when the sample is restricted to FTFY workers. After these corrections, the earnings dispersion trends for workers look remarkably similar for the two samples. Formal analysis confirms this visual impression--the estimated trends in earnings dispersion are nearly identical. Thus, restricting the samples to FTFY workers, no significant discrepancy in earnings variance is found between the two data sets: both the CPS and NLSY79 show a general trend of increasing earnings dispersion over time. The trends in earnings dispersion among non-FTFY workers, however, appear to be different in the two samples. Closer examination of the two earnings distributions shows clearly that the distribution of reported annual earnings among non-FTFY workers in the CPS is both strongly downshifted and skewed skewed curve of a usually unimodal distribution with one tail drawn out more than the other and the median will lie above or below the mean. skewed Epidemiology adjective Referring to an asymmetrical distribution of a population or of data more to the left than in the NLSY79. CPS respondents who do not work full time and year round not only report lower earnings, on average, but also the bottom tail of their distribution reaches much farther down the earnings scale. These differences already are pronounced in 1979, and they grow over time, thus contributing directly to the growing discrepancy between the two samples. For both groups of workers, annual earnings reports are higher in the NLSY79 than in the CPS by about 20 percent at the median. This begins to suggest that the primary source of the discrepancy may be underreporting in the CPS. The most likely explanation is differences in the respective questionnaires, because neither sample bias nor attrition bias Attrition bias or exclusion bias in epidemiology is a kind of selection bias caused by attrition of subjects. This can be due to:
adj. For or during less than the customary or standard time: a part-time job. part or part-year workers with irregular HEIR, IRREGULAR. In Louisiana, irregular heirs are those who are neither testamentary nor legal, and who have been established by law to take the succession. See Civ. Code of Lo. art. 874. schedules and sources of earnings. In addition, the NLSY79 is administered as a face-to-face interview, whereas the cps, except for the initial interview, usually is administered by telephone. 14 This probably will raise the validity and reliability of the NLSY79 data relative to the cps. The longitudinal basis of the NLSY79 provides a continuing relationship between the respondents and the survey organization. The promise of confidentiality has been met over time, and respondents may feel more comfortable disclosing sensitive information on earnings. Also, in the CPS, proxy reports may be a factor. All of this suggests that the discrepancies in non-FTFY annual earnings reports between the CPS and the NLSY79 may be due to underreporting in the CPS. It is worth reiterating, however, that the regression trend estimates obtained from these samples should be interpreted with care. They were found to be highly sensitive Adj. 1. highly sensitive - readily affected by various agents; "a highly sensitive explosive is easily exploded by a shock"; "a sensitive colloid is readily coagulated" to small changes in sample selection and model specification. The structure of the analytic an·a·lyt·ic or an·a·lyt·i·cal adj. 1. Of or relating to analysis or analytics. 2. Expert in or using analysis, especially one who thinks in a logical manner. 3. Psychoanalytic. question, which focuses analysis on the trends within age over time, leads to both relatively small cell sizes for estimating dispersion, and a mismatch mismatch 1. in blood transfusions and transplantation immunology, an incompatibility between potential donor and recipient. 2. one or more nucleotides in one of the double strands in a nucleic acid molecule without complementary nucleotides in the same position on the other between sample structure and the analytic task. To obtain stable estimates of the time trend, one would need relatively long periods of observation within age groups. The cohort scheme of the NLSY79, with its 8-year moving age window over time, only provides a maximum of 8 years during which any respondents are observed at a particular age, and some of the age segments include less than 2 years of observation.(15) Of course, the equivalent CPS sample reflects the same constraints CONSTRAINTS - A language for solving constraints using value inference. ["CONSTRAINTS: A Language for Expressing Almost-Hierarchical Descriptions", G.J. Sussman et al, Artif Intell 14(1):1-39 (Aug 1980)]. . While the goal of benchmarking the NLSY79 against the CPS is an important one, the NLSY79 sample structure is not ideal for answering the question posed here, and it is not clear that the survey would ever be used in this fashion. With that caveat, however, the findings described in this article still attest To solemnly declare verbally or in writing that a particular document or testimony about an event is a true and accurate representation of the facts; to bear witness to. To formally certify by a signature that the signer has been present at the execution of a particular writing so as to the validity of the NLSY79 data. Researchers should therefore take advantage of these data to examine the longitudinal questions for which this survey was designed. In general, the National Longitudinal Surveys, with their unique employer identification codes, remain the only longitudinal data set with an accurate measure of job and employer stability--a significant feature, given the many contradictory empirical findings in this field.(16) The age range covered by the survey provides a detailed window into the period when roughly two-thirds of lifetime job changes and wage growth occur.(17) These also are the formative formative /for·ma·tive/ (for´mah-tiv) concerned in the origination and development of an organism, part, or tissue. years of labor market labor market A place where labor is exchanged for wages; an LM is defined by geography, education and technical expertise, occupation, licensure or certification requirements, and job experience experience when long-term relationships with employers are established. The two National Longitudinal Survey cohorts also bracket In programming, brackets (the [ and ] characters) are used to enclose numbers and subscripts. For example, in the C statement int menustart [4] = ; the [4] indicates the number of elements in the array, and the contents are enclosed in curly braces. the growth in earnings inequality that emerged in the 1980s. Together, the cohorts of the National Longitudinal Surveys provide a unique resource for the analysis of these and other important economic and social issues covering the last 30 years. Notes (1) The Panel Study of Income Dynamics (PSID), begun in 1968, is conducted by the Survey Research Center, Institute for Social Research, University of Michigan (body, education) University of Michigan - A large cosmopolitan university in the Midwest USA. Over 50000 students are enrolled at the University of Michigan's three campuses. The students come from 50 states and over 100 foreign countries. . The PSID is a longitudinal study longitudinal study a chronological study in epidemiology which attempts to establish a relationship between an antecedent cause and a subsequent effect. See also cohort study. of a representative sample of U.S. individuals (men, women, and children) and the family units in which they reside. It emphasizes the dynamic aspects of economic and demographic behavior, but its content is broad, including sociological and psychological measures. As a consequence of low attrition rates Noun 1. attrition rate - the rate of shrinkage in size or number rate of attrition rate - a magnitude or frequency relative to a time unit; "they traveled at a rate of 55 miles per hour"; "the rate of change was faster than expected" and the success of recontact efforts, the sample size has grown dramatically in recent years, from about 7,000 core households in 1990 to almost 8,700 in 1995. As of 1995, the PSID had collected information about more than 50,000 individuals spanning as much as 28 years of their lives. For more information on the PSID, visit their website at http://www.isr.umich.edu/src/psid/. (2) The National Longitudinal Surveys (NLS), sponsored and directed by the Bureau of Labor Statistics Bureau of Labor Statistics (BLS) A research agency of the U.S. Department of Labor; it compiles statistics on hours of work, average hourly earnings, employment and unemployment, consumer prices and many other variables. , gather detailed information about the labor market experiences and other aspects of the lives of six groups of men and women. Over the years, a variety of other government agencies, such as the National Institute of Child Health and Human Development, the Department of Defense, and the Department of Education, the Department of Justice, the National Institute on Drug Abuse The National Institute on Drug Abuse (NIDA) is a United States federal-government research institute whose mission is to "lead the Nation in bringing the power of science to bear on drug abuse and addiction. , and the National School to Work Office, have funded components of the surveys that provided data relevant to their missions. As a result, the surveys include data about a wide range of events such as schooling and career transitions, marriage and fertility fertility: see infertility. fertility Ability of an individual or couple to reproduce through normal sexual activity. About 80% of healthy, fertile women are able to conceive within one year if they have intercourse regularly without contraception. , training investments, child-care usage, and drug and alcohol use. The depth and breadth of each survey allow for analysis of an expansive variety of topics such as the transition from school to work, job mobility, youth unemployment, educational attainment Educational attainment is a term commonly used by statisticans to refer to the highest degree of education an individual has completed.[1] The US Census Bureau Glossary defines educational attainment as "the highest level of education completed in terms of the and the returns to education, welfare recipiency, the impact of training, and retirement decisions. The first set of surveys, initiated in 1966, consisted of four cohorts. These four groups are referred to as the "older men," "mature women," "young men," and "young women" cohorts of the NLS, and are known collectively as the "original cohorts." In 1979, a longitudinal study of a cohort of young men and women aged 14 to 22 was begun. This sample of youth was called the National Longitudinal Survey of Youth 1979 (NLSY79). In 1986, the NLSY79 was expanded to include surveys of the children born to women in that cohort, with the new cohort called the NLSY79 Children. In 1997, the NLS program was again expanded with a new cohort of young people aged 12 to 16 as of December 31, 1996. This new cohort is the National Longitudinal Survey of Youth 1997 (NLSY97). The National Longitudinal Surveys, especially the NLSY79, have exceptional retention rates. As a result, many NLS survey members have been followed for many years, some for decades, allowing researchers to study large panels of men, women, and children over significant segments of their lives. For more information on the National Longitudinal Surveys, see the NLS Handbook
This article is about reference works. For the subnotebook computer, see .
(3) See Peter Gottschalk and Robert A. Moffitt, "Changes in the structure of earnings in three longitudinal data sets," 1997, unpublished. (4) The Current Population Survey (CPS), which uses a scientifically selected sample of about 50,000 households, is conducted monthly for the Bureau of Labor Statistics by the Bureau of the Census Noun 1. Bureau of the Census - the bureau of the Commerce Department responsible for taking the census; provides demographic information and analyses about the population of the United States Census Bureau . The CPS provides statistics on the labor force status of the civilian noninstitutional population of the United States United States, officially United States of America, republic (2005 est. pop. 295,734,000), 3,539,227 sq mi (9,166,598 sq km), North America. The United States is the world's third largest country in population and the fourth largest country in area. , aged 16 years of older. In the CPS, respondents are asked about their activity during the week that includes the 12th day of the month, the so-called reference week. As such, the CPS is a cross-sectional survey of the population, as opposed to a longitudinal survey like the NLS. For more information on the CPS, see BLS See Bureau of Labor Statistics. Handbook of Methods, Bulletin 2490 (Bureau of Labor Statistics, April 1997), pp. 4-14. (5) See Thomas MaCurdy, Thomas Mroz, and R. Mark Gritz, "An Evaluation of the National Longitudinal Survey of Youth," Journal of Human Resources The fancy word for "people." The human resources department within an organization, years ago known as the "personnel department," manages the administrative aspects of the employees. , Spring 1998, pp. 345-436. (6) To further minimize heterogeneity het·er·o·ge·ne·i·ty n. The quality or state of being heterogeneous. heterogeneity the state of being heterogeneous. , this study excludes Hispanics from the samples analyzed an·a·lyze tr.v. an·a·lyzed, an·a·lyz·ing, an·a·lyz·es 1. To examine methodically by separating into parts and studying their interrelations. 2. Chemistry To make a chemical analysis of. 3. . The study by Gottschalk and Moffitt made no such exclusion. (7) For the regression-eligible sample used here, ESR-type students represent about 15 percent of the respondents in 1979, dropping to 5 percent in 1985 and down to 1 percent by 1988. (8) See Peter J. Diggle, Kung-Yee Liang, and Scott L. Zeger, Analysis of Longitudinal Data, (New York New York, state, United States New York, Middle Atlantic state of the United States. It is bordered by Vermont, Massachusetts, Connecticut, and the Atlantic Ocean (E), New Jersey and Pennsylvania (S), Lakes Erie and Ontario and the Canadian province of , Oxford University Press), 1994. (9) See Gottschalk and Moffitt, "Changes in the structure of earnings," p. 7. (10) S-PLUS is an enhanced version of the S environment for data analysis. Unix and Windows versions See Windows. are available from MathSoft, Inc. The programs used for the analysis in this paper are available from the authors. (11) As in Chart 1, 2-year age groups are used. For FTFY workers, the values average about 180 respondents per cell for the NLSY79 and about 870 respondents per cell for the CPS. For non-FTFY workers, the corresponding values average about 90 and 300, respectively. (12) For this figure ages within a year are pooled, but the distributions have been compositionally adjusted for the differences in marginal age distributions between the CPS and NLSY79. (13) See MaCurdy and others, "An Evaluation of the National Longitudinal Survey of Youth." (14) In the CPS, respondents are part of the survey for 4 consecutive months, then they are out of the survey for the following 8 months, and finally they are back in the survey for 4 more months the following year. The first interviews are supposed to take place in person, at the home of the respondents, although face-to-face interviews are not always possible. In any case, subsequent interviews are conducted by telephone. (15) Ages 20 to 29 provide 8 years of observation each, other ages in the 16-to-36 year range provide 8 minus the difference to the closer of the two endpoints. In the analysis by Gottschalk and Moffitt, which only included up to survey year 1988, only three ages (20 to 23) would have provided 8 years of observation; all others would have provided fewer years of observation. (16) See A.D. Bernhardt, M. Handcock, and M. Scott, "Trends in Job Instability and Wages for Young Adult Men," Journal of Labor Economics The Journal of Labor Economics, published by the University of Chicago Press presents international research examining issues affecting the economy as well as social and private behavior. , Part 2, October, 1999, pp. S65-90. (17) See Kevin Murphy There are many people named Kevin Murphy:
American pathologist and bacteriologist who discovered the bacteria that causes gas gangrene. , "Empirical Age-Earnings Profiles," Journal of Labor Economics, April 1990, pp. 202-29; and Robert Topel and Michael Ward Michael Ward may refer to:
Mark S. Handcock is a statistician and Martina Morris is a sociologist at Pennsylvania State University Pennsylvania State University, main campus at University Park, State College; land-grant and state supported; coeducational; chartered 1855, opened 1859 as Farmers' High School. , and Annette Bernhardt is a sociologist at the Institute on Education and the Economy, Teachers College Columbia University Columbia University, mainly in New York City; founded 1754 as King's College by grant of King George II; first college in New York City, fifth oldest in the United States; one of the eight Ivy League institutions. |
|
||||||||||||||||||
Printer friendly
Cite/link
Email
Feedback
Reader Opinion