[ Main page of Report | Contents of Report ]
The 1996 longitudinal panel of the Survey of Income and Program Participation (SIPP), collected by the U.S. Bureau of the Census, is the primary data source that we used for examining the low-wage labor market in our study. Because of the wide range of study questions, we used different samples and methodological approaches for different types of analyses. We discuss these issues in this methodological appendix.
[ Go to Contents ]
The 1996 SIPP is a large, multipanel, longitudinal survey that collected demographic and socioeconomic information on a nationally representative sample of U.S. households. The data cover the period from late 1995 to early 2000. SIPP provides detailed monthly measures on labor force participation (for those age 15 and older), income, participation in public programs, and household composition. Our study also used data from several SIPP topical modules that contain information on supplemental topics and on sample members' experiences before the beginning of the panel period. Finally, the SIPP data were supplemented with state-level data on the economic conditions and poverty levels in the states.
The 1996 SIPP panel is particularly well suited for the study, for several reasons. First, because it covers a period between late 1995 and early 2000, we can examine the dynamics of the low-wage labor market during the post-PRWORA period. Second, because it contains detailed monthly information on jobs each sample member held during the panel period, we can conduct individual-level longitudinal analyses of employment spells and wage progression.
The SIPP data also have several advantages over other national data sets. Cross-sectional data sets, such as the March Current Population Survey (CPS), can provide point-in-time information on low-wage workers, but they do not allow analyses of individual-level employment and earnings experiences over time. The Panel Study of Income Dynamics (PSID), begun in 1968, is a longitudinal study of a representative sample of people in the United States that contains information through 1999. Thus, the PSID covers the post-PRWORA period and, because it is a long panel, has more information than SIPP on employment histories. However, because PSID data have been collected annually (and recently every other year), compared to every four months for SIPP, recall error is likely to be larger in the PSID. This is a particularly important problem for this study, because the job spells of many low-wage workers are likely to be short. Furthermore, sample sizes are much larger in SIPP (more than 40,000 households were sampled for the 1996 SIPP, whereas the 1999 PSID contains information on only about 7,000 families). The National Longitudinal Survey of Youth (NLSY) is limited to people who were ages 14 to 21 in 1978, so data from the NLSY are not well suited for examining the experiences of low-wage workers of all ages.
Adults followed in the SIPP panel come from a nationally representative sample of households in the civilian, noninstitutionalized population of the United States. Sample members were interviewed once every four months during the 48-month panel period. If original (primary) sample members older than age 14 moved from their original residences, they were interviewed at their new addresses. Secondary sample members--those who were not part of the original sample but who lived with primary sample members after the first interview--were interviewed if they were in the same household as primary sample members.
The Census Bureau used multistage sampling techniques to select a representative set of households for the 1996 SIPP panels. The first interviews for the panel began in April 1996 with a sample of 40,188 households and 95,402 primary sample members, where households in the low-income stratum were sampled at 1.66 times the rate of the higher-income stratum. (1) Sample households were divided into four "rotation groups" of roughly equal size, and one rotation group was interviewed each month. Thus, each household was interviewed in four-month intervals, called "waves." The 1996 SIPP contains 12 waves, which provide 48 months of data for each person in the sample.(2)
At each interview, sample members provided information about their experiences during the preceding four-month period, called the "reference period." For example, people in rotation group 1 whose wave 1 interviews were conducted in April 1996 (the earliest interviews) were asked about their experiences between December 1995 and March 1996. Similarly, people in rotation group 4 whose wave 12 interviews were conducted in March 2000 (the latest interviews) were asked about their experiences between November 1999 and February 2000. Thus, the 12 reference periods for the 1996 SIPP panel cover December 1995 through February 2000.
The 1996 SIPP interviews were administered using computer-assisted interviewing (CAI) to increase data quality. CAI, used for the first time in the 1996 SIPP, permitted automatic consistency checks of reported data during the interview and allowed for the use of prior-wave data for editing missing data.
The SIPP questionnaire is made up of the core questions and the topical modules. The core questions provide information on (1) demographic characteristics; and (2) work behavior, income, and program participation for each of the four months preceding the interview date. The core questions were asked in every wave interview. Sample members were asked the topical module questions after the core questions. The content of the topical module changed from wave to wave. For our purposes, the topical modules administered in wave 1 are of special interest, because they contain information on respondents' prepanel experiences (see Section 4 below).
The Census Bureau constructed a full-panel, longitudinal research file by linking the data collected for each sample person over the life of the panel. Unlike the individual core wave files that contain one record per person-month, the longitudinal file contains one record per person. The longitudinal sample that this research file represents consists of all primary sample members who have complete data (either reported or imputed) for every month of the panel (excluding months of ineligibility). This longitudinal sample contains 55,484 people and is the main sample that was used for the analysis.
The 1996 longitudinal file contains a smaller percentage of all primary sample members than in previous SIPP panels, for several reasons. First, sample attrition was higher in the 1996 panel than in earlier panels because the 1996 panel was longer (12 waves, compared to 8 waves in previous panels). For example, the sample loss rate was 35.5 percent by the end of wave 12 in the 1996 panel, but it was 26.9 percent by the end of wave 8 in the 1993 panel.(3) Second, in creating the final data files, the Census Bureau typically performs imputations for missing responses to individual questions or to entire wave interviews (see U.S. Census Bureau 2003, SIPP Data Editing and Imputation), thereby increasing the sample size in the analysis files. In creating the 1996 SIPP data files currently available, however, the Census Bureau has performed fewer imputations than in previous panels.(4)
The longitudinal research file is available online using the FERRET system. As the Census Bureau specifies, however, this system is efficient (practical) only for downloading a small number of variables, because variable requests must be performed separately for each variable using a series of menus and because downloading even a few variables takes considerable time. Our study employs a large number of variables, so we did not use the FERRET system to obtain the longitudinal data needed for the analysis.
Instead, we downloaded (from the SIPP Web page) the entire ASCII database for each of the 12 individual core wave files and constructed our own longitudinal file following the same procedures the Census Bureau used to construct its longitudinal file. Specifically, we "flattened" each core file to obtain one record per person (rather than per person-month) and merged these 12 flattened files using the unique person identification code (LGTKEY). We compared key selected variables (such as earnings and hourly wage rates) in our constructed longitudinal file to those in the longitudinal file on the FERRET system and found the variables to be identical in both data files.
Finally, to take into account nonresponse, sample attrition, and the complex sample design of the 1996 SIPP (including the oversampling of poor households), the longitudinal research file contains panel weights (which we downloaded using the FERRET system). These weights make the SIPP longitudinal sample representative of the noninstitutionalized, resident population of the United States as of March 1996 (the only month common to all four rotation groups in wave 1).(5) We used weights throughout the statistical analyses and adjusted the standard errors of our estimates to account for design effects due to weighting and clustering.
The topical modules contain more detailed information on particular topics than are contained in the core files. We used data from the topical modules to construct explanatory variables for the multivariate analysis.
The wave 1 topical module contains retrospective information on sample members' prepanel activities and experiences. The most important such information for this study concerns prepanel employment experiences (including the number of years the respondent worked at least six months, breaks from the labor force, the date last worked, and whether the respondent generally worked 35 or more hours per week since he or she first started working at least six months per year). Unlike previous SIPP panels, information on the starting dates for those in the middle of job spells at the start of the SIPP panel period (that is, who have left-censored job spells) are in the wave 1 core file and not in the topical module.
Several topical modules contain information on work schedules and health status. Data on work schedules are contained in the wave 4 and wave 10 topical modules. However, as discussed later, most of our analysis was conducted using samples of workers who began low-wage jobs at the start of the panel period (that is, in waves 1 and 2), and our analysis described the characteristics of low-wage workers and their jobs at the start of these jobs. Thus, the data on work schedules was collected too late to be useful for our study, so we did not use them in the analysis. For a similar reason, we did not use the detailed information on functional limitations and disabilities contained in topical modules 5 and 11.(6) However, we did use in the analysis the health status variable contained in each core data file concerning whether the respondent had a physical, mental, or other health condition that limited the kind or amount of work that could be done.
The state-level data for our analysis included information on states' economic conditions. We merged this state-level information by month or year (depending on data availability) to the SIPP data file using monthly (annual) information on the state in which each sample member lived.(7) We used this information to explore the relationship between state characteristics and the dynamics of the low-wage labor market in the multivariate analysis.
We used variables from the following categories of state economic indicator variables that are intended to proxy for the labor market situation faced by SIPP sample members:
Rural population share
Although we initially included all these measures as explanatory variables in our multivariate models, we ultimately narrowed the list because of the high correlation among many of the state-level measures. This high degree of multicollinearity increased the standard errors of all parameter estimates and made it difficult to isolate the separate effects of each of the state-level measures. The final list of explanatory variables included (1) the unemployment rate measures, (2) the poverty rate measure, (3) the 20th percentile of monthly wages, and (4) the rural population share.
[ Go to Contents ]
A central analysis issue for the study is how to define low-wage workers. As discussed in detail in Chapter II, researchers have used a variety of definitions of the low-wage labor market, and each definition has advantages and disadvantages. Because of project budget constraints, it was not feasible to conduct analyses using each of these measures. Therefore, we needed to select among the alternative measures.
Our primary approach for defining low-wage workers was to use the hourly wage at which a full-time worker would have annual earnings below poverty for a family of four. We calculated separate low-wage cutoff values for each calendar year the SIPP panel covered. We then classified a worker as "low-wage" if the worker's wage rate was less than the cutoff level in the calendar year when the wage rate was reported. Using U.S. Department of Health and Human Services poverty guidelines and assuming a full-time worker works 2,080 hours per year, we set the low-wage cutoff at $7.50 in 1996, $7.72 in 1997, $7.91 in 1998, $8.03 in 1999, and $8.20 in 2000. We also defined medium-wage workers as those with wage rates between one and two times the low-wage cutoff value and higher-wage workers as those with wages more than twice the low-wage cutoff value.
We adopted the absolute low-wage cutoff approach so that the analysis could focus on low-wage workers and their jobs based on a well-defined cutoff value. We did not use the minimum wage as the absolute wage cutoff value, because it sets the bar too low for defining the low-wage labor market. We rejected using definitions based on family income levels, because that approach would be appropriate for examining working poor households rather than low-wage workers.
We used the absolute wage cutoff rather than a relative wage cutoff, because the relative wage cutoff allows for no change over time in the fraction of the labor force that is defined as low wage, even if living standards of low-income workers change. For example, under the relative wage approach, a worker earning a wage rate at the 20th percentile of the wage distribution at two time points would be classified as a low-wage worker at each point, even if the wage distribution for low-wage workers shifted over time (that is, even if the worker's wage rate changed). Thus, the relative wage approach would provide less information than the absolute wage cutoff approach on the extent to which low-wage workers enter and exit the low-wage labor market over time. Furthermore, we rejected using a definition based on the skill levels of workers, because not all workers in the low-wage labor market have low skills.
We did, however, construct samples of low-wage workers using alternative definitions when we estimated the size of the low-wage labor market as part of the descriptive analysis presented in Chapter III. The rest of the analysis, however, was conducted using only the absolute low-wage cutoff measure.
Finally, one implication with the absolute low-wage cutoff measure is that the low-wage threshold was constructed for a household of average size and, thus, may be too low for larger-than-average households and too high for smaller-than-average ones (although it is correct on average). One approach for addressing this issue would be to define wage cutoff levels by household size so that the cutoff values would be higher in larger households than in smaller ones. We rejected this option, however, for two main reasons. First, the unit of analysis is the low-wage worker, rather than the low-income working household; thus, it is preferable to use a uniform definition for all workers. Second, household size often changes over time, so people's cutoff values would often change over time, which would lead to analytic complications. For example, suppose a worker held the same job and received the same wage rate in two successive months. If the worker's household size decreased in the second month, then the worker could be classified as a low-wage worker in the first month but not in the subsequent one.
[ Go to Contents ]
Our study seeks to address a broad range of research questions related to the low-wage labor market, including questions that require the analysis of employment-related data at a point in time and over specific intervals. Furthermore, to address some questions, the individual is the unit of analysis; to address others, the low-wage job or employment spell is the unit of analysis. Thus, we employed various analysis samples and statistical methods for the study.
In this section, we first discuss general issues about which workers were included in the empirical analysis and the construction of hourly wage rates. Then, we provide an overview of specific analytic issues separately by type of analysis. We provide a more complete discussion of these issues in each of the relevant topical chapters presented in this report.
Our analysis was conducted using employed SIPP sample members who were between ages 16 and 64 and who were not enrolled in school at the start of their jobs. We excluded students and older workers, because their labor market experiences are likely to be very different from those of the population that is the focus of this study.
Our analysis included information on those who worked for employers (that is, those who held jobs) and on those who owned businesses. At each wave, SIPP contains information on up to two jobs held by sample members and two businesses owned by sample members during the reference (four-month) period. Although the studies of low-wage workers reviewed in Chapter II typically examined those in jobs only, we included both jobs and businesses in our analysis, because a significant percentage of those with businesses were low-wage workers. For example, in March 1996, about 12 percent of all low-wage workers in our sample owned businesses, and an equal share of those with jobs and businesses were low-wage workers. Thus, we did not want to exclude from the analysis self-employed workers who constitute an important segment of the low-wage labor market.
For each month of the panel period, we constructed hourly wages for each job and business using detailed employment information in SIPP. SIPP contains direct information on hourly wage rates for the 60 percent of jobholders who could provide wage data in this way. Hourly wage rate information, however, is not available for the remaining 40 percent of jobholders and for all those with businesses. For these workers, we constructed hourly wages by combining information on monthly earnings (which are reported for each month of the panel period) and usual hours worked per week at each job or business during the reference period (topcoded at 84 hours), and assuming that the worker was employed for the entire month.(8) The "earnings-based" hourly wage measure was then constructed for each month by dividing monthly earnings by the number of hours worked in the month.(9)
Our preliminary analysis of the SIPP data showed that hourly wage rates fluctuated considerably over time, and especially for the constructed earnings-based measures. These fluctuations are often due to sudden large changes in wage rates that appear to be due more to reporting errors or SIPP data errors than to real wage changes. Furthermore, they yield more worker transitions into and out of the low-wage labor market than we deem plausible. Consequently, we used several methods to "smooth" the hourly wage rates to identify those who were truly in low-wage jobs:
Finally, for those with multiple jobs and businesses in a particular month, we selected the hourly wage from the job or business in which the sample member worked the most hours. In March 1996, about 11 percent of workers held multiple jobs and businesses. Thus, we defined whether a worker was a low- medium-, or high-wage worker using the wage on the selected "main" job or business in that month.
Our analysis addresses questions in four topical areas: (1) the characteristics of low-wage workers and their jobs, (2) the employment experiences of low-wage workers over a three-year follow-up period, (3) the wage growth of low-wage workers over a three-year period, and (4) the duration of low-wage job and employment spells and types of exits from the low-wage labor market. Next, we briefly provide an overview, by topical area, of the analysis samples and methodological approaches used in the study, as well as the subgroups for which separate estimates were obtained. We provide additional details in the report chapters that present the study findings. We begin here, however, with a brief discussion of general analytic issues that pertain to all analyses.
a. General Analytic Issues
Our descriptive and multivariate analyses were conducted separately for males and females, because of differences in labor market participation decisions and experiences by gender. Within each gender group, we calculated statistics for the full sample, as well as for key subgroups defined by worker and job characteristics. We used sample weights in all analyses (either the longitudinal or calendar year weights, depending on the analysis) to make our findings representative of all workers nationally.
An important component of our analysis was to compare the characteristics and labor market experiences of low-wage workers to those of medium- and high-wage workers (labeled hereafter as "higher-wage workers"). We conducted these analyses to provide a context from which to understand the findings for those in the low-wage labor market. Thus, in selected analyses, we computed statistics for workers in each of the three wage categories. For example, to help interpret findings on the percentage of time that low-wage workers were employed during the follow-up period, we also computed these employment measures for medium- and high-wage workers.
b. Describing Low-Wage Workers and Their Jobs
The main analysis sample that we used in our descriptive analysis to examine
the prevalence of low-wage jobs and the characteristics of low-wage workers
and their jobs is a cross-sectional sample of workers in March 1996. We selected
March 1996 as the reference point because
(1) it is the earliest month in the SIPP data that is covered for all sample
members; (2) the 1996 calendar year weight is constructed to make the sample
representative of the U.S. population in March 1, 1996; (3) previous
cross-sectional studies examining the low-wage sector have used the March
CPS data, so we can compare our results to those from previous studies; and
(4) the sample used to examine the overall employment experiences and wage
growth of low-wage workers was based on those who started low-wage jobs early
in the panel period. We also constructed cross-section samples of workers
in March 1997, March 1998, and March 1999 to examine changes in the prevalence
and profiles of low-wage workers over time, due to changing economic conditions
and TANF program parameters.
We described three main aspects of low-wage workers and the types of jobs they hold separately for men and women. First, we examined the fraction of all workers who are in the low-wage labor market. Second, we examined their demographic characteristics and compared them with those of medium- and high-wage workers. Finally, we described the job and other employment-related characteristics of low-wage workers.
A worker was defined as a low-, medium, or high-wage worker on the basis of the worker's hourly wage measure (on the main job or business) at the time the worker entered the sample (for example, March 1996). Similarly, worker and job characteristics were defined at the sampling point.
c. Examining Overall Employment Experiences
The analysis of the overall employment experiences of low-wage workers was conducted using only those who started jobs or businesses during the first six months of the panel period, to ensure a sufficient follow-up period for examining overall employment patterns and adequate sample sizes. We identified the first job that the worker held during the six-month period, and if the sample member held multiple jobs or businesses at the same time, we selected the job or business at which the sample member worked the most hours. We classified a sample member as a low-, medium-, or high-wage worker on the basis of the worker's average hourly wage during the month of job start and the subsequent six months (for those months in which the worker was employed). We used this six-month period to help identify "true" low-wage workers from those who held low-wage jobs for only a very short time due to temporary changes in earnings or labor supply effort, or to data errors. The follow-up period was measured at the start of the initial job, and was 42 months for all sample members (the longest period that could be examined for those who started jobs in panel month 6). Thus, the follow-up period was not measured in calendar time, but in the number of months since job start.
We constructed the following categories of outcome measures for the analysis:
We calculated summary statistics for each outcome measure for the whole sample and for selected subgroups. In addition, we conducted selected analyses using medium- and high-wage workers to place the findings for low-wage workers in perspective. All estimates were constructed using the longitudinal panel weights. We also estimated multivariate regression models to examine factors associated with positive overall employment outcomes during the follow-up period. This analysis allowed us to more efficiently examine a larger set of factors than could be examined in the descriptive analysis. The analysis also allowed us to isolate the contribution of each factor from others. In Chapter IV, we discuss the specific dependent and explanatory variables included in the models and the statistical techniques used to estimate the models.
d. Examining Wage Progression
For the analysis of wage progression, we examined the extent to which the wages of low-wage workers grow over time and what factors are associated with wage growth over a three-year follow-up period. Similar to the overall employment analysis, the wage progression analysis was conducted using only those who started jobs or businesses during the first six months of the panel period. The key difference between the wage progression analysis and the overall employment analysis is that the wage progression analysis focused on continuous measures of wage growth, whereas the overall employment analysis focused on employment patterns over the follow-up period.
As described earlier, to classify job starters as low-, medium-, or high-wage workers, we based our initial classification of workers into these three groups based on their average wages during the first six-month period after they started their jobs. Categorizing people into low-, medium-, or high-wage workers at any given point in time has two potential issues especially important for the wage growth analysis. First, if a worker misreports his or her wages at the time of job start, we may incorrectly classify an individual into a wage type that may not be their real wage type. Second, people sometimes obtain jobs that may not be related to their true ability levels and may soon move into a job that more closely matches their true human capital level. For example, if a worker with low productivity gets a high-wage job, he or she may not be able to sustain that job for long and may soon move into a low-wage one. Conversely, a high-productivity worker may have found a low-wage job and might soon move to a higher-wage job (defined as a medium- or high-wage job). Both these factors work in the direction of potentially large wage growth for low-wage workers (or lower wage growth for high-wage workers), especially in the early periods after job start. We were particularly concerned about minimizing the effects of any data errors, as these errors do not reflect true changes in wages. Thus, as described earlier, we smoothed wages and took the six-month average of wages after job start to classify workers into wage categories.(11) (We call this initial period to classify workers into wage categories "period 0.") While this smoothing is likely to reduce the noise due to data errors to a large extent, residual errors could still remain, and we may be overstating wages for low-wage workers. Consequently, in our analysis examining wage growth over time, we start with the average wage in the first six-month period after the period we used to define their initial worker type and examine their wage growth over the following three-year period (period 1 through period 6). For trends in wages over time, we present average wages of those employed in period 1, average wages of those employed in period 2, average wages of those employed in period 3, and so on. For the analysis of individual workers' wage growth over time, we compare wages and job characteristics of those workers who were employed in both the first and last periods (i.e., period 1 and period 6) regardless of their employment in other periods.
Additionally, the wage progression analysis was conducted using only those who were employed at various follow-up points. This is because hourly wages are observed only for those who were employed--they are not observed for nonworkers (they are missing, not zero).(12) Thus, the sample to examine wage progression was restricted to those who reported being employed in various follow-up intervals (six-month intervals), so that initial hourly wages could be compared to hourly wages reported later. This is the usual approach used in the literature to address wage growth issues.(13)
We measured wage progression as the difference in (real) hourly wage rates at various fixed time points after the start of the low-wage job. We also measured wage growth as the percentage increase in real wages (relative to the starting wage) at the same follow-up points. In addition, we constructed indicator variables signifying whether the worker's wage increased, decreased, or stayed the same (and by how much).
We conducted descriptive and multivariate analyses to examine wage progression. To help interpret the wage growth results, we also compared the distribution of key job characteristics for the initial low-wage job and the most recent job held in the last year of the panel period. This analysis provides information on whether potential increases in wage growth between years 1 and 4 are associated with improvements in other job characteristics, such as the availability of fringe benefits, hours worked, and occupations. As discussed in more detail in Chapter V, we also conducted multivariate analyses to examine factors associated with wage progression.
e. Examining Spell Durations and Types of Exits
An important component of our analysis was to examine the distribution of the length of continuous job and employment spells for low-wage workers and the extent to which these spells end in higher-wage jobs or in nonemployment. This duration analysis differed from (but complements) the overall employment analysis in several respects. First, the duration analysis focused on the low-wage spell rather than the low-wage worker. Thus, the analysis file for the duration analysis contains one record per spell month rather than one record per person. Second, the duration analysis focused on the length of continuous low-wage job and employment spells, whereas the overall analysis described patterns of potentially discontinuous employment and nonemployment spells that workers experienced over a fixed follow-up period.
A central, and complicated, analytic issue is how to define job, employment, and nonemployment spells. To facilitate this discussion, we first list the five possible states into which a low-wage worker could exit:
Using these possible exit states, we conducted duration analyses for four types of job and employment spells, each of which addresses a slightly different analytic question:
Similar procedures were used to construct spells for those who began medium- and high-wage jobs during the panel period.
We examined also two types of spells for our analyses of reentry into the low-wage labor market. First, we examined the rate at which those who exited their low-wage jobs into nonemployment (that is, into exit states 4 and 5) returned to the low-wage and higher-wage labor markets. Second, we examined the extent to which those who exited their low-wage jobs into higher-wage jobs returned to the low-wage sector.
In sum, the samples for the duration analysis included entry cohorts of job, employment, and nonemployment spells that began during the panel period. Job and employment spells were classified as low-wage (or higher-wage) on the basis of the hourly wage rate at the start of the spell, and a spell ended if the worker exited into one of the various exit states described above. These samples allow us to answer such hypothetical questions as (1) Of those who begin a low-wage job, what percentage will still be working at that job one year later? and (2) Of those who begin a low-wage job, how many will leave that job and go directly into a higher-paying job? Similarly, the sample for the analysis of nonemployment spells allows us to answer such questions as: Of those who exit a low-wage job into nonemployment, how many will become reemployed in low-wage or higher-wage jobs within eight months?
We used standard life table statistical methods to estimate the proportion of spells that ended within a given number of months after the start of the spell (that is, cumulative exit rates). As discussed in Chapter VI, these methods adjust for right-censored spells (that is, spells still in progress at the end of the panel period) and left-censored spells (spells in progress at the start of the panel period). We conducted analyses for the full sample of males and females, as well as for key population subgroups defined by worker characteristics at the start of the spell. We also compared findings for low-wage workers to those of medium- and higher-wage ones.
f. Subgroup Analysis and Sample Sizes
As discussed, we conducted all analyses separately for male and female workers because of differences in labor market participation decisions by gender. In addition, within each gender group, we conducted selected analyses for key subgroups of low-wage workers defined by their demographic and job characteristics at the start of their low-wage jobs (for the overall employment, wage progression, and duration analyses). The subgroup analysis provides information on whether labor market experiences differ for different groups of low-wage workers. We selected the following policy-relevant categories of subgroups across whom we hypothesized study findings might differ:
Table A.1 displays subgroup definitions and sample sizes by type of analysis. In addition to these subgroups, we examined the relationship between a broader set of characteristics and key labor market outcomes in our multivariate analysis (as discussed further in the main report).
| Subgroup | Describing Demographic and Job Characteristics(a) | Overall Employment Analysis(b) | Wage Progression Analysis(c) | Employment Spell Duration Analysis(d) | ||||
|---|---|---|---|---|---|---|---|---|
| Males | Females | Males | Females | Males | Females | Male Spells | Female Spells | |
| Total | 3,466 | 5,044 | 522 | 817 | 491 | 693 | 8,274 | 11,133 |
| Individual and Household Characteristics | ||||||||
| Age (in Years) | ||||||||
| Younger than 20 | 172 | 177 | 67 | 56 | 61 | 47 | 613 | 538 |
| 20 to 29 | 1,106 | 1,256 | 198 | 262 | 189 | 225 | 2,246 | 2,721 |
| 30 to 39 | 941 | 1,476 | 122 | 240 | 127 | 197 | 2,174 | 3,118 |
| 40 to 49 | 687 | 1,227 | 71 | 157 | 69 | 145 | 1,791 | 2,811 |
| 50 to 59 | 462 | 709 | 39 | 81 | 45 | 79 | 1,147 | 1,726 |
| 60 or older | 118 | 199 | 25 | 21 | 303 | 419 | ||
| Race/Ethnicity | ||||||||
| White and other non-Hispanic | 2,401 | 3,777 | 379 | 614 | 357 | 523 | 6,047 | 8,418 |
| Black, non-Hispanic | 441 | 726 | 62 | 103 | 51 | 86 | 993 | 1,639 |
| Hispanic | 644 | 541 | 81 | 100 | 83 | 84 | 1,234 | 1,276 |
| Educational Attainment | ||||||||
| Less than high school/GED | 811 | 890 | 147 | 155 | 131 | 118 | 1,754 | 1,844 |
| High school/GED | 1,460 | 2,254 | 212 | 344 | 198 | 300 | 3,281 | 4,667 |
| Some college | 586 | 871 | 84 | 138 | 112 | 204 | 1,454 | 2,109 |
| College graduate or more | 629 | 1,029 | 79 | 180 | 50 | 71 | 1,785 | 2,713 |
| Has a Physical, Mental, or Other Health Condition That Limited the Kind or Amount of Work That Could Be Done | ||||||||
| Yes | 300 | 454 | 61 | 83 | 50 | 60 | 624 | 848 |
| No | 3,186 | 4,590 | 461 | 734 | 441 | 633 | 7,650 | 10,485 |
| Household Income as a Percentage of the Poverty Level | ||||||||
| 100 percent or less | 514 | 665 | 120 | 179 | 121 | 153 | 1,161 | 1,564 |
| 101 to 200 percent | 1,137 | 1,380 | 150 | 247 | 137 | 199 | 2,340 | 2,905 |
| More than 200 percent | 1,835 | 2,999 | 252 | 391 | 233 | 341 | 4,773 | 6,864 |
| Household Type | ||||||||
| Single parent with children | 331 | 958 | 60 | 203 | 58 | 173 | 706 | 2,160 |
| Married couple with children | 1,333 | 1,931 | 204 | 313 | 203 | 268 | 3,412 | 4,541 |
| Married couple without children | 864 | 1,220 | 129 | 165 | 112 | 139 | 2,288 | 2,695 |
| Other adults without children | 958 | 935 | 129 | 136 | 118 | 113 | 1,868 | 1,937 |
| Job Characteristics | ||||||||
| Hourly Wages | ||||||||
| Less than $5.00 | 886 | 1,383 | 130 | 257 | 114 | 187 | 1,410 | 1,967 |
| $5.00 to $6.00 | 816 | 1,299 | 130 | 250 | 136 | 223 | 1,790 | 2,917 |
| $6.00 to $7.00 | 984 | 1,378 | 158 | 208 | 122 | 153 | 2,222 | 2,971 |
| $7.00 to $7.50 | 800 | 984 | 104 | 102 | 119 | 130 | 2,852 | 3,478 |
| Hours Worked per Week | ||||||||
| 1 to 19 | 99 | 459 | 40 | 128 | 40 | 113 | 339 | 1,314 |
| 20 to 34 | 435 | 1,252 | 100 | 252 | 82 | 209 | 1,005 | 2,913 |
| 35 to 40 | 1,750 | 2,633 | 280 | 371 | 263 | 318 | 4,149 | 5,767 |
| More than 40 | 1,202 | 700 | 102 | 66 | 106 | 53 | 2,781 | 1,339 |
| Weekly Earnings | ||||||||
| Less than $150 | 532 | 1,470 | 110 | 332 | 108 | 280 | 1,154 | 3,199 |
| $150 to $299 | 2,216 | 3,183 | 342 | 454 | 291 | 356 | 4,626 | 6,566 |
| $300 to $600 | 738 | 391 | 70 | 31 | 92 | 57 | 2,494 | 1,568 |
| Occupation | ||||||||
| Professional/technical | 487 | 719 | 36 | 70 | 39 | 72 | 1,258 | 1,897 |
| Sales/retail | 396 | 784 | 57 | 136 | 55 | 119 | 1,001 | 1,901 |
| Administrative support/clerical | 174 | 999 | 33 | 159 | 31 | 132 | 450 | 2,391 |
| Service professions/handlers/cleaners | 1,008 | 1,805 | 187 | 324 | 167 | 267 | 2,227 | 3,619 |
| Machinists/construction/production/ transportation |
1,131 | 681 | 151 | 104 | 143 | 83 | 2,681 | 1,329 |
| Farm/agriculture/other workers | 290 | 56 | 58 | 24 | 56 | 20 | 657 | 196 |
| Health Insurance Coverage(e) | ||||||||
| Yes | 1,823 | 3,350 | 196 | 403 | 179 | 343 | 4,851 | 7,410 |
| No | 1,663 | 1,694 | 326 | 414 | 312 | 350 | 3,433 | 3,923 |
| Source: 1996 SIPP files. Note: All samples exclude those in school and workers younger than age 16 and older than age 64 at the start of their jobs. a. This sample includes low-wage workers in March 1996 with a positive 1996 calendar year weight. b. This sample includes workers who (1) started low-wage jobs during the first six months of the panel period, (2) who have a positive longitudinal panel weight, and (3) had at least 38 months of follow-up data. c. This sample includes workers who (1) started low-wage jobs during the first six months of the panel period, (2) were employed at some point between 2.5 and 3 years later, and (3) had a positive longitudinal panel weight. d. This sample includes low-wage employment spells that started during the panel period or were in progress at the start of the panel period (about 20 percent of spells are left-censored). The sample includes the spells of only those with positive longitudinal panel weights. A worker can contribute more than one spell to the sample. e. These figures pertain to health insurance coverage from all sources, including coverage through the employer as well as from other sources. We used this variable instead of the employer-based health insurance coverage variable, because data on overall health insurance coverage is available monthly, whereas the employer-based coverage variable pertains only to jobs in progress at the time of the interview. Thus, the employer-based health insurance variable could not always be linked to the job under investigation, which led to a significant number of missing values. However, the subsets of health insurance variables overlap considerably: the source of health insurance coverage was the employer for 80 percent of those with any coverage. |
||||||||
[ Go to Contents ]
(1) The sample size for the 1996 SIPP was larger than for previous panels. For example, the 1990 panel contains 21,900 sampled households and 43,799 sampled people.
(2) The 1996 SIPP redesign called for 12 panels, rather than the 8 used in previous SIPP panels.
(3) The sample loss rate at the end of wave 8 in the 1996 panel was 32.8 percent.
(4) The Census Bureau has performed imputations for "Type Z" noninterviews, which occurred when an interview was conducted with at least one household member but not with one or more sample people in the household. The Census Bureau, however, has indicated that it will not perform additional imputations for the 1996 SIPP panel.
(5) The longitudinal data file also contains calendar year weights. The 1997, 1998, and 1999 calendar year weights pertain to the January population in those years. The 1996 weight, however, pertains to the March population.
(6) Topical modules 3, 6, 9, and 12 contain detailed information on medical expenses and the utilization of health care, but these data cover topics that are beyond the scope of the analysis.
(7) For nine states with relatively few SIPP respondents, the data do not identify the state individually, but rather in three groups: (1) Maine and Vermont; (2) Iowa, North Dakota, and South Dakota; and (3) Alaska, Idaho, Montana, and Wyoming. For these groups, we inserted mean characteristics across all states in the group.
(8) Initially, we used job and business start and end dates to calculate the exact number of weeks that the worker was employed in the month. However, we found in the data that workers who started jobs in the middle of the month tended to report monthly earnings for the full month. For example, we found many instances where workers reported the same monthly earnings in months in which they worked only part of the month (that is, in months when they started their jobs) and in subsequent months. Thus, in order to avoid inflating the constructed wage rates, we assumed that workers were employed for the full month.
(9) We did not use the earnings-based measure for those who directly reported an hourly wage, because we believe that the direct measure is more accurate. This approach has typically been used in the literature discussed in Chapter II.
(10) These categories are not mutually exclusive.
(11) As noted in Chapter II, the usual extent of data cleaning performed in earlier SIPP waves was not done for the 1996 longitudinal files.
(12) According to economic theory, an individual chooses not to work if the person's market (offered) wage is lower than the person's reservation wage (the minimum wage for which the person would be willing to work). Otherwise, the individual chooses to work, and hours worked are adjusted to equate the reservation and market wages. Thus, for nonworkers, the reservation wage is missing, and one cannot assign a zero wage rate to these individuals.
(13) As discussed in Chapter VI, we examined the extent of potential sample selection biases in our estimates by comparing the characteristics of workers who were included and excluded from the analysis sample. The results from this wage growth analysis may represent a best-case scenario, because the sample is likely to overrepresent those who had positive employment outcomes at the various points.
(14) For the overall employment and wage progression analyses, the hourly wage rate and weekly earnings subgroups were formed using the average wage during the month of job start and the subsequent six months.
Main Page of Report | Contents of Report
Home Pages:
Human Services Policy
(HSP)
Assistant Secretary for Planning and Evaluation
(ASPE)
U.S. Department of Health and Human Services
(HHS)