Atypical work and pay.

1. Introduction

   Even as the economy continues to recover, the U.S. is increasingly
   becoming a nation of part-timers and free-lancers, of temps and
   independent contractors. This "disposable" labor force is the most
   important trend in business today, and it is fundamentally changing
   the relationship between Americans and their jobs. For companies
   large and small, the phenomenon provides a way to remain globally
   competitive while avoiding the vagaries of market cycles ... But for
   workers, it can mean an end to security and sense of significance
   that come from being a loyal employee. One by one, the tangible
   bonds that once defined work in America are giving way.

   Castro (1993, p. 43.)

The frequency of alternative work arrangements such as temporary employment, on-call work, and independent contracting--nonstandard forms of employment commonly referred to as "atypical work"--has steadily increased in recent decades (see, for example, Segal and Sullivan 1997). Research on atypical work/alternative work arrangements--we shall use the two terms interchangeably--has tended to focus on the nature and extent of these arrangements and their impact on a worker's employment history (e.g., Addison and Surfield 2006). Although compensation levels have been accorded less attention, the thrust of much of the earnings literature is pessimistic: Compared with regular or open-ended employment, alternative work arrangements often appear to offer inferior remuneration. The public perception remains that workers in these alternative work arrangements constitute a disposable workforce characterized by low-paying and precarious or unstable jobs.

Abstracting for the moment from the complications introduced by the diversity of alternative work arrangements, there are a number of reasons why their incumbents might earn less than regular workers. For example, wage differentials between atypical workers and those in open-ended employment can arise from supply-side differences among heterogeneous workers and employer preferences. On the supply side, a subset of workers may favor nonstandard work forms over regular employment, and be prepared to accept lower pay. One example would be adults who are heavily involved in household production, and another is older workers who have moved out of career jobs. But supply-side differences alone are insufficient to sustain a permanent ceteris paribus differential in favor of regular workers. In addition, atypical workers and regular workers must not be fully fungible and employers must not be indifferent between them.

Negative wage gaps for atypical work are not the only possibility admitted by the competitive model. Atypical workers may earn higher wages than their counterparts in regular employment if the aggregate demand for these nonstandard arrangements is greater than the number of individuals seeking such employment. Thus, to adapt Hirsch's (2005, p. 527) argument in respect of part-timers, if firms use atypical work as low-cost means of adjusting to variable and uncertain demand, the relatively large supply of atypical workers required may run up against mobility or substitution constraints--across labor markets delineated by geography, occupation, and industry--and give rise to positive equilibrium wage gaps for atypical workers.

Negative wage gaps may of course also be more apparent than real. For example, worker ability and employment in an alternative work arrangement might be negatively correlated. In this case, an unfavorable wage gap will be attributable not to type of work arrangement but, rather, to lesser ability. We would certainly like to control for ability so as to avoid drawing faulty inferences about earnings and work arrangement in the presence of such sorting behavior on the part of low-ability individuals.

This pattern of negative and positive wage gaps may largely reflect the nature of temporary work. Consider first agency temporaries, namely, workers paid by a temporary help agency. Such workers are typically hired by a client firm where the work is expected to be temporary or is of uncertain duration. Companies may find it cost-effective to hire such workers, even if they are of lesser productivity than regular workers, because they are less costly to terminate (Autor 2003). They may thus form a buffer stock. It is likely that the employment of both agency temps and another type of atypical worker, direct-hire temporaries, will increase relative to regular employment in an expansion. In addition, companies may use agency temps to fill vacancies until permanent hires are made and, in some cases, to recruit permanent workers from the ranks of the agency workers themselves. In the latter case, for those choosing to take a low-paid agency job to queue for regular employment, we will observe a negative equilibrium wage gap. Temporary help agencies probably enjoy scale economies in recruiting and screening of temporary and permanent workers. They pool jobs across companies and can provide workers with a better selection of schedules than can any company acting on its own, while offering employment continuity (albeit with the agency). Similar considerations apply with respect to permanent positions, especially in tight labor markets. Tightening labor markets also encourage the substitution of agency workers for in-house, on-call worker pools and the direct-hire temporaries.

Why might companies pay agency temporaries less than regular workers, that is, discriminate in favor of the latter? (1) In a labor market consisting of good and risky workers (who may be either good or bad but share poor work histories), firms will be wary of offering permanent contracts to workers who may turn out to be lemons. The firm's choice of the two types of worker depends on the costs per unit of output for each type. Houseman, Kalleberg, and Erickcek (2003) argue that temporary agencies can reduce the costs to firms of hiring risky workers, and may be a more efficient solution to the informational problem than the probationary contract. For example, if agencies are more adept at screening and matching workers, their use raises the probability that the worker will prove suitable--and cheaper to hire and fire if unsuitable (Booth, Francesconi, and Frank 2002). The use by the client company of agencies to hire risky workers allows it to expand the supply of labor and deflect the need to raise wages for new and incumbent workers.

Again, it is not automatically the case that temporary agency workers will experience lower wages. For example, in the recovery phase a firm will have to pay higher wages to new hires. The higher wages may be generalized throughout the workforce to avoid adverse morale and productivity effects. This threat may be deflected through hiring the new workers, who make more money than the rest, through agencies. On the assumption that agency workers and regular workers are homogeneous, Houseman, Kalleberg, and Erickcek (2003) show that employers may discriminate in favor of agency workers, thereby raising the wages of new entrants without raising the wages of existing employees. The maintained hypothesis is that regular workers are, for a variety of reasons, less knowledgeable about agency temps' wages than they are of other workers' wage levels.

These arguments suggest that temporary agency workers can earn both negative and positive wage differentials. As a practical matter, however, it seems likely that agency temporaries will be less likely to enjoy a premium (and more likely to experience a negative differential) because of their less favorable skill composition than other atypical worker groups, such as contracting company employees (and independent contractors). For example, in their study of hospital use of agency temporaries, Houseman, Kalleberg, and Erickcek (2003, p. 111) find that the cases in which companies use temporary workers to buy time to recruit permanent employees at lower wages are precisely those "where workers must meet clear educational or certification requirements to perform jobs and where the costs of having an unqualified person staff a position are high."

Small and medium-sized firms will contract with outside 'knowledge workers' because it is not cost-effective to provide the service in-house. For example, economies of scale in the provision of, say, computer support activities might lead the firm to rely on the resources and experience of a computer services contracting company (Abraham and Taylor 1996). It has also been argued that client companies of company contract workers will also include those who are "unable or unwilling to satisfy the demands of knowledge workers for increased control over their work," both in terms of its execution and timing, leading such workers to turn to atypical contracting arrangements (Forde and Slater, 2005, p. 254). Expert workers in such high-skill professional, managerial, scientific, and technical occupations may be expected to earn higher wages than the generality of regular workers by virtue of their greater skill endowments. It is not clear that they will enjoy an additional premium, outside of the situation discussed earlier, and situations in which companies are using higher-priced contract workers to buy themselves time to recruit regular workers and hence discriminate in favor of these atypical workers. That said, in cases where the skills of contract workers considerably exceed those of regular workers, it is likely that a positive ceteris paribus wage gap will be recorded, especially in circumstances where, as here, we are not controlling for fringes.

The main contribution of the present paper is to estimate wage differences for the full set of alternative work arrangements while simultaneously controlling for observed demographic characteristics and unobserved person-specific fixed effects. The wage gaps that we identify are the starting point for separate lines of inquiry. We justify our more restricted focus on the grounds that there has been inadequate analysis of earnings differences between atypical and regular workers--as was also the case until very recently for part-time workers (see Hirsch 2005).

Existing work on atypical worker compensation is either dated (in cross section) or overly restricted in the range of alternative wage arrangements examined (in longitudinal data). Accordingly, we seek to update previous findings using subsequent waves of the main data set (the Contingent and Alternative Employment Arrangement Supplement or CAEAS) available to researchers in this area (see section 2), while using a different longitudinal data set (the National Longitudinal Survey of Youth, 1979 Cohort or NLSY79) to correct for permanent unobserved differences between individuals for a wider range of alternative work arrangements than hitherto.

A chronological review of the literature on atypical work and its remuneration is provided in Table 1. Many of the studies are descriptive. The study by Cohany (1998) in row 6 of the table is representative in this regard. It charts little change in the characteristics of those employed in such activities over time (specifically, between the first and the second CAEAS) and, in comparing their median weekly earnings with those of workers in open-ended employment, reports that agency temporaries and on-call workers fare particularly poorly; for example, the median weekly earnings of agency temporary workers are only two thirds of those of regular workers. That said, and again consistent with the findings from the first CAEAS, independent contractors and contract workers enjoy a wage gap of between 15% and 21%.

Differences in wages across the various employment arrangements will reflect differences in worker characteristics. Thus, for example, Cohany reports that atypical workers are younger, have lower educational attainments, and are more concentrated in the lower-paid industries and occupations than are regular workers. The study by Polivka, Cohany, and Hipple (2000) in row 8 adopts a multivariate analysis to control for such differences. The authors' ordinary least squares (OLS) wage regressions put the wage disadvantage of agency temporary work at -5% to -9%. (2) Much of the difference in wages between temporary agency work and regular work appears therefore to be explained by differences in worker characteristics. On the other hand, these authors find no such attenuation for those workers at the opposite end of the earnings spectrum: Contractors are still estimated to enjoy a wage premium of 23%, at least in 1997, on par with raw results of Cohany.

Polivka, Cohany, and Hipple (2000, p. 77) recognize the limitations of the data set, noting that compensation is influenced by a number of variables not contained in the CAEAS, such as "firm-specific factors, personal tastes, and other unobserved differences that might influence who is in these arrangements." The analyses of Segal and Sullivan (1997, 1998), summarized in rows 4 and 7 of Table 1, in part anticipated these concerns. Although limited to a comparison of agency temporary workers with their counterparts in regular employment, these two studies were the first to control for individual unobserved heterogeneity in atypical worker earnings determination. The study in row 7 of the table is notable for its use of administrative data (rather than the CAEAS), extracting a 10-year sample panel from the 1984-1994 quarterly records contained in the Washington State Unemployment Insurance system, identifying temporary employees through their industry affiliation. (3) A comparison of the authors' pooled and fixed effects regression estimates indicates that accounting for permanent unobserved individual heterogeneity reduces the earnings deficit of temporary workers by between 10% and 15%.

In the final study discussed here of single-parent female welfare recipients initially obtaining atypical work in the temporary help service sector versus other industries, Heinrich, Mueser, and Troske (2005) (row 10) report that this choice does not prejudice their future earnings development or continued employment--or, for that matter, welfare recidivism. Welfare recipients beginning work in this sector do earn substantially less than their counterparts in other sectors, but this difference does not seem to be the result of unmeasured characteristics. Moreover, the low earnings are not permanent: After two years the differences between those initially in atypical work are virtually the same as their counterparts who had jobs in other industries. This faster earnings growth is shown to be partly the result of atypical workers moving to other higher-paying industries. In addition, there is no difference in the proportions of workers who do not have a job one year later across industries, including temporary help. The bottom line from this study is that wage gaps may not persist and that welfare recipients obtain opportunities for future advancement by working in the temporary help service sector.

In the present study, we seek to build and improve upon the existing literature in two ways. First, as noted earlier, unlike the more sophisticated of the ceteris paribus studies in rows 4, 7, and 10 of Table 1, each of our two data sets allows us to examine a wider set of alternative work arrangements than agency temporary employment alone--even if we cannot easily investigate the persistence of the wage gap. In addition, our longitudinal data set offers a wider array of controls than available in the Current Population Survey (CPS), including proxies for worker ability and a measure of actual experience, while we can update the cross section data set with additional rounds of the CAEAS for 1999 and 2001. Our second contribution to the analysis of the effect of alternative work arrangement on pay is technical in nature. Polivka, Cohany, and Hipple (2000) elect not to use the natural log of a worker's hourly earnings as their dependent variable given certain distributional aspects of the data. More precisely, the right-skewness in the earnings of independent contractors prevented them from directly estimating the Mincerian semi-logarithmic earnings function that has found such strong support in the literature. We adopt a technique that takes the variances of the earnings attaching to the various work forms into consideration, without abandoning the conventional Mincerian earnings function.

Our analysis proceeds as follows. We first outline the empirical model. The two data sets used in this inquiry are next introduced. Then our detailed findings are presented, sequenced by data set. A brief interpretative section concludes.

2. The Empirical Model

We use three linear estimating techniques to uncover the effect of atypical work on earnings. For our cross section data set, we estimate a simple ordinary least squares model having pooled the cross section (CAEAS) data. Both random effects and fixed effects least squares models are used to analyze the longitudinal (NLSY79) data.

Consider the underlying wage determination model that includes worker ability:

E([w.sub.i,t]|[x.sub.i,t], [AWA.sub.i,t],[c.sub.i]) = [[beta]'[x.sub.i,t] + [delta] [AWA.sub.i,t] + [c.sub.i], t = 1, 2, ..., T, (1)

where [w.sub.i,t] is the (log) wage earned by worker i at time t, [x.sub.i,t] are the corresponding observed worker characteristics, [AWA.sub.i,t] is a dummy variable equal to one if a worker i is engaged in an alternative work arrangement at time t (zero otherwise), and [c.sub.i] is worker ability. The parameter [delta] is the wage gap or differential that is associated with employment in an alternative work arrangement.

When Equation 1 is estimated by OLS, we have

[w.sub.i,t] = [[beta]'[x.sub.i,t] + [delta][AWA.sub.i,t] + [c.sub.i] + [u.sub.i,t]. (2)

Absent controls for worker ability, OLS will estimate

[w.sub.i,t] = [[beta]'[x.sub.i,t] + [delta][AWA.sub.i,t] + [v.sub.i,t], (3)

where [v.sub.i,t]=([c.sub.i]+[u.sub.i,t]). Equation 3 will still provide unbiased estimates of [delta] provided E[[v.sub.i,t]|[[x.sub.i,t]]=0.

However, we will still have to correct the estimates obtained from Equation 3 to allow for differences in the wage variances observed across the various alternative work arrangements. The conventional method of obtaining the percentage impact of atypical work on wages is to exponentiate the estimated coefficient, [delta], and then subtract one. This approach will yield biased and inconsistent estimates of the differential in percentage terms in the presence of heteroskedasticity. For dichotomous variables, such as the various atypical work arrangements, differences in the dispersion of the dependent variable (in this case the natural log of the wage rate) may correlate with the error term. Depending on the direction of the correlation, the conventional method of obtaining the percentage differential will either over- or underestimate the true wage effect of atypical employment. This concern has been widely cited in the health economics literature (e.g., Manning 1998) and there is a significant concern that this is the case in our present analysis. Our ceteris paribus estimates are corrected using a procedure adopted in Blackburn (2007), the technical details of which are remitted to Appendix A.

Related to Equation 3 is the random effects linear estimator. It allows us to aggregate our three waves of longitudinal data into one pooled sample, assuming the cross-sectional differential has not changed over time. Although this estimator does not take unobserved individual heterogeneity into account, it will produce unbiased and consistent coefficient estimates by allowing for the cross-correlation in the error term that arises when we have repeated observations on the same individual.

If, as may be hypothesized, ability and employment arrangements are negatively correlated, however, the estimate of the wage differential will be biased downward. We can remove worker ability from the model using a fixed effects specification that will also be estimated alongside Equation 3. The fixed effects specification allows for not only individual-specific intercepts but also year-specific intercepts, as follows

[w.sub.i,t] = [[alpha].sub.t] + [[phi].sub.i] + [[beta]'[x.sub.i,t] + [delta][AWA.sub.i,t] + [u.sub.i,t], (4)

where [[alpha].sub.t] captures the impact, if any, that time has on worker earnings and where the individual-specific intercept [[phi].sub.i] controls for any time-invariant unobserved worker characteristics, such as ability. Any elements of [x.sub.i,t] that are unchanging over time are omitted from Equation 4.

3. The Data

We use two main data sets to estimate the differential attaching to atypical work: the CAEAS to the CPS--as well as the parent CPS itself--and the NLSY79. The advantage of the former data set is its size, given the relatively small proportion of workers in certain alternative work arrangements. The disadvantage is that there is no overlap of households across the supplements, ruling out panel estimation methods. Each attribute is reversed in the case of the latter data set.

The CAEAS/CPS Wage Data

We first extracted one sample from each of the four biennial CAEAS supplements. To allow for the inclusion of the 2001 CAEAS (4) and to increase the number of regular workers included in our analysis, we link each February CAEAS to the subsequent March basic CPS survey using the matching algorithm outlined in Madrian and Lefgren (1999). The CPS collects information on industry, occupation, and wages only for those regular workers who are in an outgoing rotation group. The CAEAS collects these data for atypical workers without regard to their rotational status. We link the two surveys to allow for the inclusion of regular workers for whom the crucial wage data is elicited in either February or March. For those regular workers who are observed as being in the March outgoing rotation group, we include only those who are observed in the February CAEAS/CPS data, who report having the same employer (and being employed) in both months, and who have the same activities and occupation in each month. The four cross sections were subsequently pooled to obtain more precise estimates of the wage differentials attaching to the different work arrangements (results for the separate cross sections are available from the authors upon request), while accounting for year effects. Note that the four samples included only those individuals who were employed in the week prior to their February interview. This restriction was imposed because those recorded as unemployed, or as non-participants, would not report any labor force or wage data.

The wage variable is the hourly wage rate. It is constructed following the detailed procedure outlined in Polivka (1999). Workers either reporting or who were calculated to earn less than two dollars an hour or more than $150 an hour were excluded from the samples. Further, only those individuals aged 25 to 65 years at the time of the February survey were included to minimize the problem of compounding differential supply responses. Individuals with incomplete demographic, industrial, and occupational data were excised.

Workers are classified into seven mutually exclusive work arrangements. The first two categories pertain to open-ended employment and comprise regular workers and screened workers. Following the convention established in the literature, we distinguish between five types of atypical workers: agency temporary workers, direct-hire temporary workers, on-call workers, contract company workers, and independent contractors.

Regular workers are those individuals who are directly hired into an open-ended employment arrangement using standard interviewing methods, rather than having first been screened through an alternative work arrangement. We classify individuals as screened workers if they are currently engaged in open-ended employment and they indicate that immediately prior to being employed in this capacity they were employed by the firm in an alternative work arrangement (i.e., without any break in employment continuity). We distinguish between the two types of open-ended employment to allow for the possibility that initially serving an employer as an atypical worker strengthens the bond between employer and employee, and thereby influences the wage paid to such workers.

Since we have already mentioned the five atypical work categories--agency temps, direct-hire temporary workers, on-call workers, contract workers, and independent contractors--only parenthetical remarks are in order. First, since agency temps are defined as those receiving their pay from a temporary help agency, they will include regular employees of the agency in addition to agency temps proper. However, as noted by Houseman and Polivka (2000), a 1989 Industry Wage Survey indicates that these workers comprise only 3.2% of an agency's total employment. Second, as regards contract workers (who differ from independent contractors in that they, like their agency temporary brethren, rely on a third party to provide them with the necessary clients or projects), we require that they have only one client and usually work at that client's workplace. Finally, we note that the CAEAS does not include a specific question on direct-hire temporaries. After Polivka, Cohany, and Hipple (2000, p. 43), we classified individuals as direct-hire temporaries if they indicated that their job was temporary and not intermediated, or that they could not stay in their job as long as they wished for other reasons (they were working only until the completion of a specific project, were hired for a fixed interval, expected to work for less than a year because the job was temporary, etc.).

The NLSY79 Wage Data

The National Longitudinal Survey of Youth, 1979 Cohort (NLSY79) is the product of repeated interviews with individuals aged 14-21 years at the time of the initial interview in 1979 (and therefore just beginning to enter the labor market). To run simple OLS regressions, we first extracted three different cross sections from the data set, covering employees in employment in 1994, 1996, and 1998. With one exception the restrictions were the same as those applied to the CAEAS/CPS: We exclude from the NLSY79 samples those workers who had accumulated less than nine weeks' tenure with their current employer because the survey does not elicit information such as industry affiliation or the worker's occupational classification from the respondent unless he or she meets this particular service threshold. The wage rate is measured on an hourly basis, but is constructed from information on earnings and usual hours worked relevant to the frequency at which the individual is actually paid. Next, for random and fixed effects estimates, an unbalanced panel was constructed, covering all three years and applying the same restrictions in respect to missing data and the truncation of the wage distribution. This unbalanced panel construct does not require that workers be recorded as employed for all three years, only for at least two of the three years. It was designed to permit a more precise estimate of the wage gap by including as many observations on individuals as possible.

Individuals were initially grouped into six possible job categories rather than the eight constructed from the CAEAS/CPS. The first two categories of regular workers and screened workers are identical and, as before, the distinction is based on the notion that previously screened workers may be in a better job match. The same is true for agency temporaries and direct-hire temporaries. Since we are unable to distinguish between the two types of contract work, we will speak of a composite contractors/consultants group. The remaining category is a catch-all, comprising 'all other' work types, of which the most numerous subgroup is self-employment.

4. Findings

Descriptive wage data for the four CAEAS/CPS cross sections and the three waves of NLSY79 data are given in Tables 2 and 3, respectively. Beginning with the larger CAEAS/CPS samples, we see that agency temporaries, direct-hire temporaries, and on-call workers earn lower wages than do regular workers. Of the five atypical work arrangements in the CAEAS, agency temps clearly fare the worst, with earnings that are 14% to 32% lower than open-ended employment. Direct-hire temporaries earn between 94 and 98 cents on the regular dollar. And for on-call workers, the raw wage differential is between -6% and -13%. At the other end of the wage spectrum are contract workers and independent contractors. Contract workers earn 6% to 15% more than regular workers, while the wage premium attaching to independent contracting ranges from 21% to 29%. If anything, screened workers earn somewhat higher wages than regular workers.

The earnings of atypical workers vis-a-vis regular workers are directionally the same in the NLSY79 data, but as shown in Table 3 the differences are systematically sharper. For example, the negative wage differential associated with agency temporary employment now ranges from 36% to 43%, whereas workers in contracting/consulting earn a premium of between 22% and 44%. There is less evidence in the NLSY79 that screened workers enjoy any wage advantage over regular employees; indeed, in two out of three waves they earn (4%) less.

Work diaries maintained by the NLSY79 respondents provide us with some additional human capital controls not found in the CAEAS/CPS data. Specifically, we have direct data on a worker's (cumulative) general labor market experience as well as his or her tenure on the current job. The NLSY79 gives the actual number of weeks that a respondent has been employed since entering the survey, as well as the actual number of weeks employed with the current employer (or employment type in the case of contractors/consultants and other work types). Moreover, we can also form a standardized measure of the number of jobs held by individuals by dividing the reported total number of jobs held by (cumulative) general labor market experience. This jobs measure can be also viewed as an inverse proxy for the attractiveness of the worker to an employer. Descriptive information on each of these variables is provided in Table 4 for the 1994 wave of the NLSY79. Not surprisingly, workers in atypical work arrangements have substantially less tenure with their (agency) employer than do regular workers. In terms of general labor market experience, it would appear that contractors/ consultants have been employed slightly longer than regular workers. And, despite their having spent fewer years in employment, agency temps have held more jobs on average than those engaged in open-ended employment.

We can also compute tenure--but not experience and the number of jobs held--for the CAEAS/CPS. As for the NLSY79, tenure is with the current employer for regular and screened employees and with the agency or contracting intermediary for agency temps and contractors, respectively. For the separately identified category of independent contractors--and on-call workers--tenure pertains to the employment arrangement rather than the employer. Comparing the two panels of Table 4, it is apparent that tenure is somewhat longer in the CAEAS/CPS data set.

We now turn to the multivariate cross-section results, beginning with those for the larger CAEAS/CPS samples. In all cases, we report results for pooled data--rather than the four separate cross sections--since there is no overlap across the supplements. (Separate results for the cross sections are available from the authors upon request.) Table 5 presents results for the simple semi-logarithmic wage model, and entered for reference purposes only since, as was noted earlier, the coefficient estimates produced by this model may not yield consistent estimates of the actual wage differential in percentage terms. Specifically, for dichotomous variables, such as the five atypical work arrangements, differences in the dispersion of the dependent variable (in this case the natural log of the wage rate) may correlate with the model's error term--even if there is no such difficulty with the continuous regressors.

We therefore offer heteroskedasticity-corrected estimates in Table 6, using the procedure outlined in Blackburn (2007). To obtain the differential in percentage terms, one now simply has to multiply the point estimates in the table by 100. Also note that, as in the previous table, separate results are presented for samples that either include or exclude workers with allocated earnings. As shown by Hirsch and Schumacher (2004), the use of allocated earnings may cause the estimates of the differential to be biased toward zero. When the CPS allocates earnings to a non-respondent, it does so using a number of characteristics among which the more important are age, gender, education, and occupation. Each worker with missing wage information is matched to an individual or 'donor' with identical characteristics. Although these variables are good predictors of individual earnings, the authors show coefficient estimates for those characteristics that the CPS does not rely upon to allocate earnings--such as atypical work--will be biased in analyzing workers with allocated wages. For example, given the fairly small number of workers engaged in on-call work, we will understate the (negative) effect that this work form has on earnings given the high probability that the CPS will allocate to a non-respondent on-call worker the wage rate earned by a regular worker.

Returning therefore to the heteroskedasticity-corrected estimates in Table 6, we first consider results for the larger sample; that is, including workers with allocated earnings. (Recall that the coefficient estimates for continuous variables simply carry over from Table 5.) (5) From the first column of the table, we see that controlling for worker characteristics preserves the wage penalty attaching to direct-hire temporary employment (at about 6%) while attenuating that to agency temporary work (now 15%). Note that the F-tests at the foot of Table 5 reject the possibility that agency and direct-hire temping have the same implications for worker earnings. As regards to the other atypical work arrangements, it can be seen that most of the raw wage disadvantage (from 6% to 13% in Table 2) disappears in the ceteris paribus estimates: On-call workers now receive 1.1% less than regular workers. We also observe attenuation in the wage premiums attaching to each type of contract work: Electing to work as a contract worker (independent contractor) now serves to increase the worker's earnings by 12 (16)%.

The effect of including a quadratic term in tenure is shown in the second column of Table 6. The inclusion of the tenure arguments further reduces the wage disadvantage of agency and direct-hire temporary workers and the modest wage disadvantage of on-call workers is eliminated. For contract workers and independent contractors, however, the wage premium is largely unaffected at 14% and 15%, respectively. (The impact of the linear term in tenure on wages is some 2.4% a year, as shown in Table 5.)

Much the same pattern of results is obtained once we remove workers with allocated earnings from the sample. The results are given in columns 3 and 4, which otherwise correspond to the specifications in columns 1 and 2, respectively. There is little suggestion that allocation materially affects the results for agency and direct-hire temps or on-call workers, although for each type of contract work the positive premiums increase.

Two final observations should be made with respect to the CAEAS/CPS sample. First, we fail to find any evidence that being screened initially as an atypical worker before being offered regular employment has any effect on the worker's wage. Second, the notion of there being a composite atypical work category is contraindicated: The hypothesis tests at the foot of Table 5 confirm that different work arrangements play distinct roles in earnings determination.

Turning next to the analysis of the NLSY79 earnings data, which are unaffected by any earnings imputation problem, our random effects estimates are presented in Table 7. (Results for individual waves of the NLSY79 are available from the authors upon request.) Column 1 presents results for the simple semi-logarithmic specification, while column 2 applies the correction for heteroskedasticity suggested by Blackburn (2007). (6) Columns 3 through 5 introduce controls for ability. Beginning with the results in the first two columns of the table, it can be seen that correcting for heteroskedasticity considerably elevates the absolute magnitude of the wage differential attaching to atypical work. This is most obviously the case for agency temps and contractors/consultants (as well as other work types composite that we will not subsequently discuss given its heterogeneity). The negative wage differential associated with agency temporary employment is some 30% and statistically significant, although it is somewhat reduced vis-a-vis the raw differentials earlier reported in Table 3 that ranged from -36% to -44%. The large wage premium enjoyed by contractors/consultants remains very strong: such workers earn wages that are 30% higher than regular employment. As far as the other work forms are concerned, there is every indication that the negative differential observed in the raw data is an artifact of the type of workers filling these positions, while in the case of screened workers we now see for the first time some indication that screening leads to higher wages in open-ended employment.

As was the case for the CAEAS/CPS sample, we find no support for the use of an aggregated temporary work category. Not surprisingly, the results provided at the foot of Table 7 suggest that the two types of temporary employment have a significantly different effect on worker earnings. We also find that aggregating over the various atypical work categories is inappropriate.

The performance of the human capital variables is broadly as expected. Thus, greater firm-specific human capital, as measured by tenure, significantly increases earnings: Each year spent with an employer adds to a worker's earnings by about 3%. And the number of jobs held by an individual negatively impacts earnings: Each additional job that a worker has held since entering the labor market reduces earnings by about 3%. However, the coefficient estimate for general labor market experience is only marginally significant.

Unique to the NLSY79 is the availability of information on the respondent's Armed Forces Qualification Test (AFQT) test scores. The AFQT is a set of standardized tests used by the military to assess the abilities and knowledge of recruits in the areas of general science, arithmetic reasoning, word knowledge, paragraph comprehension, numerical operations, coding ability, auto and shop knowledge, knowledge of mathematics, mechanical comprehension, and electronics information. We formed four separate categories for worker ability. Thus, for example, we combined the scores on the word knowledge and paragraph comprehension tests. The resulting aggregate score was then regressed on a set of age and education dummies at the time the test was administered, which were allowed to have nonlinear effects. The residuals from this regression were used as a proxy for a worker's verbal ability in the OLS wage regression. Measures of mathematical ability, coding ability, and the catch-all of practical ability were derived in a similar manner. (7)

The results of the OLS regressions including these observed ability measures are given in the third column of Table 7. For expositional convenience, we report only the heteroskedasticity-corrected estimates. We see that of the four ability proxies, only coding and math ability are associated with increases in earnings. The coefficient estimates for the other two ability proxies are of the expected sign but fail to achieve significance at conventional levels. Inclusion of the ability proxies does not materially alter the point estimates attaching to agency temporary employment (cf. column 2). However, we do find some evidence that ability and employment as a direct-hire temp are negatively correlated. That said, and contrary to our expectations, the strong wage premium for contractors/consultants is further elevated once we control for ability.

The last two columns of Table 7 provide disaggregated results by gender. The results for males are given in column 4 and those for females in column 5. It can be seen that male agency temps have a wage penalty of 45.1%, direct-hire temps a premium of 6.4%, and contractors/ consultants a premium of 32.8%. The corresponding differentials without the ability proxies--not shown in the table--are -44.0%, 7.4%, and 32.8%, respectively. If there is seemingly scant evidence that the ability measures influence the male point estimates, the position is somewhat different for female differentials given in column 5. Thus the wage disadvantage of female agency temps of 13.3% is lower in absolute terms than the corresponding estimate without the ability controls (-15.3%). A similar result obtains for direct-hire temps where the reported small premium of 1.1% is to be compared with a without-ability differential of -1.1%. If ability and temporary work are negatively correlated, failure to take ability into consideration will overstate the negative differential. Finally, in the case of female contractors/consultants, the with- and without-ability premia are 62.6% and 66.5%, respectively. In short, for females, introduction of the ability proxies serves to diminish the absolute value of the wage differential, consistent with ability and work arrangement being positively correlated in this case.

In Table 8, we report results of a more conventional accounting for unobserved worker heterogeneity, including ability, using a fixed effects estimator. Results are presented for the full sample and for males and females separately. For the all-worker sample we observe a virtual halving of the wage disadvantage of agency temping to around -16.5% (compare this value with the estimate in column 2 of Table 7). It also emerges that employment as a direct-hire temporary worker gives higher rewards than regular employment after controlling for unobserved worker heterogeneity. Both results would support the hypothesis that ability and employment in a temporary arrangement are negatively correlated, leading to overestimates of the (negative) effects of these work forms on earnings. By the same token, we observe an attenuation of the wage premium enjoyed by contractors/consultants: they are now recorded as earning a premium of almost 20% versus 30% in the random effects specification in column 2 of Table 7. Seemingly higher-ability workers are sorting themselves into contracting employment, although there was no indication of this when using the proxy for ability (see columns 2 and 3 of Table 7).

In line with the results obtained for the full sample, the results for males in the second column of Table 8 point to a further reduction in the wage penalty for agency temporary employment after allowing for unobserved individual heterogeneity, although the differential remains a sizeable 34%. For direct-hire temporaries, the wage premium increases sharply, which is again consistent with lower ability workers locating jobs in this sector. Interestingly, there is almost no attenuation in the wage premium for contracting/consulting work over regular employment, which is steady at around 30%.

The results for females in the final columns of the table are rather different from those of males. The negative and positive differentials found for female temporary workers and female contractors/consultants, respectively, in the baseline random effects specification are reversed in the fixed effects estimation. It seems that female temporary workers have unobserved characteristics that lead to lower earnings, and conversely for female contractors/consultants. In the former case, these characteristics most likely include a preference for flexible employment by reason of family or household demands as well as marginal attachment to the labor force. The fixed effects results would suggest that this group of females actually fares better when employed in temporary jobs than regular work. As far as female contractors/consultants are concerned, the modest negative differential observed in fixed effects can be taken as more indicative of their being of higher ability. If higher-ability females are disproportionately opting for contracting/consulting work, then one would expect to find an upward bias in the cross-section estimate of the wage effect.

The bottom line from the NLSY79 analysis is that the use of AFQT and fixed effects produce results that are broadly in the same direction, most obviously in the case of females. The results are much weaker when using AFQT scores to proxy for ability than modeling unobserved individual heterogeneity formally. The standardized tests do not cover certain key aspects of worker heterogeneity (such as motivation and labor force attachment) even if they do seem to do more than confirm the individual's ability to take the relevant tests.

5. Conclusions

Our analysis extends prior research into the effects of atypical employment on compensation in a number of directions. We have provided updated cross-section evidence (incorporating two additional rounds of the CAEAS). We have provided panel estimates for a full set of atypical work arrangements. We also have corrected the earnings function estimates for skewness in the earnings distribution throughout, while adjusting for allocated workers in the CPS. With the exception of direct hire temporary workers, who have lower ceteris paribus relative earnings in the CAEAS/CPS and conversely in the NLSY79, our two data sets produce directionally similar results. We refer in particular to the wage disadvantage of agency temps and the wage premium of contractors/consultants. The absolute differentials are stronger in the NLSY79. However, when we restricted the CAEAS/CPS sample to individuals aged 28 to 34 years, consistent with the former sample, we found that differentials increased in absolute value in five out of six cases (these results are available from the authors upon request).

The wage disadvantage of agency temps and the wage premium of contractors/ consultants diminish when we formally allow for unobserved individual heterogeneity. Agency temps have a differential of about 17% and contractors a premium of 20%. But these values are in turn influenced by gender differences. We find that males continue to suffer a rather substantial wage penalty in respect of agency temping (-34%) and a sizeable premium (31%) for contract/consulting work. For females, however, the fixed effect estimates point to a solid premium in agency temping (15%) and a modest negative differential (3.2%) for contracting/ consulting. Selection therefore plays a major role in the case of females.

Finally, to the extent that our results offer a more optimistic evaluation of atypical work--to include a wage premium for direct-hire temping and the appearance of a modest premium for screened workers--than the balance of the literature and, indeed, run counter to the thrust of much of the public debate concerning these alternative wage arrangements, we must enter the caveat that we do not here consider fringes, such as access to health and pension benefits. It has been reported that atypical workers have less access to and lower coverage of health insurance than their counterparts in regular employment (e.g., Polivka, Cohany, and Hipple 2000). How do these associations qualify interpretation of our results? In the case of contractors, it may be the case that the earnings premium detected here is sufficient to compensate them for the lack of coverage. By the same token, such individuals may be more likely to have spouses that are covered, so that we may expect a purposive sorting of workers into contracting/consulting, especially those with a stronger preference for money than benefits. Admittedly, agency temps present a different picture. Some such individuals are likely to be temps force majeure and their health insurance coverage a matter of concern. On the other hand, others will choose to be temps because they prefer the flexibility. Arguably, the latter are much more likely to be covered by their spouse's health insurance. We propose to investigate these issues in future research, although even the better data set in this regard (the CAEAS/CPS) offers an imperfect platform for this inquiry.

Appendix A

The standard convention in calculating the ceteris paribus wage differential involves the regression of the log wage on a vector of worker characteristics, x. In using the log wage, rather than the wage in level terms, we are assuming that the underlying wage determination model is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (A1)

where [w.sub.i] is the wage rate earned by worker i, [x.sub.i] is a vector of worker observed characteristics, and [AWA.sub.i] is a variable equal to one if the worker is engaged in an alternative work arrangement (zero otherwise). This model has strong theoretical support, and is usually transformed to a model of the form

log ([w.sub.i]) = [x'.sub.i] [beta] + [psi] [AWA.sub.i] + [[epsilon].sub.i], (A2)

where [[epsilon].sub.i] is a standard error term. OLS estimation provides consistent estimates of the log-wage parameters [beta] and [psi] if E[[[epsilon].sub.i]|[x.sub.i], [AWA.sub.i]=0. However, it may not yield consistent estimates of the percentage effect of atypical work on wages.

After estimating (A2), it is conventional to obtain the differential attaching to, say, atypical employment via

[[delta].sub.AWA] = [[epsilon].sup.[spi]] - 1. (A3)

Blackburn (2007) argues that this approach is flawed if

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (A4)

In particular, if the error terms are normal and there is heteroskedasticity in the error terms that is associated with employment in an AWA, then the conventional calculation of the differential will provide an inconsistent estimate of the wage impact of AWAs. This possibility can be tested by taking the residuals from the estimation of Equation A2, squaring them, and then regressing the squared terms on the same elements of [x.sub.i] and [AWA.sub.i] as follows

[e.sup.2.sub.i] = [x'.sub.i] [[lambda].sub.1] + [[lambda].sub.2] AWA. + [u.sub.i] (A5)

If [[lambda].sub.2] does not equal zero, then (A3) will not yield the true percentage impact of AWAs on wages; specifically, a positive (negative) [[lambda].sub.2] would imply that any estimates of [[delta].sub.AWA] using (A3) are over- (under-) stated.

In these circumstances, Blackburn shows that a more appropriate estimate of the percentage impact of AWAs on wages, [[DELTA].sub.AWA], is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (A6)

where the sampling variance of [[DELTA].sub.AWA], is computed as

var([DELTA].sub.AWA]) = [[([DELTA].sub.AWA] + 1).sup.2] * var([psi])] + 1/2 [[([DELTA].sub.AWA] + 1).sup.2] x var([[lambda].sub.AWA])]. (A7)

Appendix B shows the results of fitting Equation A5 for both data sets. It indicates that the use of (A6) is required.

Appendix B
Ordinary Least Squares Estimates of Error Dependence
(Dependent Variable: Squared Error Term)

                              Estimated Error
                                 Dependence
Category                      (Standard Error)

Pooled CASAS/CPS data
  Screened workers           -0.022 ** (0.009)
  On-call workers             0.157 *** (0.010)
  Agency temporaries          0.048 *** (0.013)
  Contract workers            0.051 *** (0.016)
  Direct-hire temporaries     0.092 *** (0.006)
  Independent contractors     0.299 *** (0.005)

Random effects NLSY79 data
  Screened workers            0.011 (0.011)
  Contractors/consultants     0.075 * (0.043)
  Agency temporaries         -0.113 *** (0.036)
  Other work types            0.117 ** (0.049)
  Direct-hire temporaries     0.002 (0.038)

* p < 0.1.

** p < 0.05.

*** p < 0.01.

Appendix C
Full Ordinary Least Squares Regression Estimates of Atypical Worker
Wage Differentials, Pooled CAEAS/CPS Data (Dependent Variable:
Log Hourly Wage)

                                         Estimate
Variable                             (Standard Error)

Worker category
  Screened workers                      0.012 (0.008)
  Direct-hire temporaries              -0.072 *** (0.007)
  Contract workers                      0.101 *** (0.020)
  Agency temporaries                   -0.109 *** (0.015)
  On-call workers                      -0.052 *** (0.013)
  Independent contractors              -0.016 ** (0.008)

Demographics
  Tenure (in years)                     0.024 *** (0.001)
  [Tenure.sup.2]/100                   -0.041 *** (0.002)
  Age                                   0.028 *** (0.001)
  [Age.sup.2]                          -0.000 *** (0.000)
  Black                                -0.080 *** (0.006)
  Other ethnicity                      -0.032 *** (0.008)
  Female                               -0.118 *** (0.006)
  Married                               0.086 *** (0.005)
  Married females                      -0.094 *** (0.007)

Education level completed
  High school diploma                   0.163 *** (0.006)
  Some college                          0.237 *** (0.007)
  Associate's degree                    0.294 *** (0.008)
  Bachelor's degree                     0.408 *** (0.008)
  Master's degree                       0.515 *** (0.010)
  JD/MD/PhD                             0.570 *** (0.015)

Place of residence
  Urban                                 0.133 *** (0.004)
  North Central                        -0.054 *** (0.005)
  South                                -0.080 *** (0.005)
  West                                 -0.006 (0.005)

Industry
  Construction/mining                   0.194 *** (0.017)
  Transportation, communications,       0.234 *** (0.017)
  and utilities
  Finance, insurance,                   0.174 *** (0.017)
  and real estate
  Manufacturing                         0.180 *** (0.016)
  Retail/wholesale trade               -0.025 (0.016)
  Personal services                     0.005 (0.019)
  Public administration                 0.208 *** (0.017)

Skill category
  Business services                     0.123 *** (0.018)
  Professional services                 0.064 *** (0.016)
  Technical/sales workers              -0.097 *** (0.007)
  Service workers                      -0.371 *** (0.007)
  Skilled workers                      -0.141 *** (0.007)
  Clerical workers                     -0.241 *** (0.006)
  Operators/laborers                   -0.296 *** (0.007)

Year dummy
  1997                                  0.048 *** (0.004)
  1999                                  0.123 *** (0.004)
  2001                                  0.203 *** (0.006)

N                                  73,610
Adjusted [R.sup.2]                      0.36

* p < 0.1.

** p < 0.05.

*** p < 0.01.

Appendix D
Full Regression Estimates: Ordinary Least Squares Cross-Section
Regression Estimates of Atypical Worker Wage Differentials,
NLSY79 Data (Dependent Variable: Log Hourly Wage)

                                           Estimate
Variable                               (Standard Error)

Worker category
  Screened workers                      -0.011 (0.011)
  Direct-hire temporaries                0.002 (0.038)
  Contractors/consultants                0.075 * (0.043)
  Agency temporaries                    -0.113 *** (0.036)
  Other work types                       0.117 ** (0.049)

Experience level
  Jobs (standardized)                   -0.033 *** (0.009)
  Experience (in years)                  0.010 * (0.005)
  [Experience.sup.2]                     0.001 (0.000)
  Tenure (in years)                      0.028 *** (0.002)
  [Tenure.sup.2]                        -0.001 (0.000)

Demographics
  Age                                    0.050 ** (0.023)
  [Age.sup.2]                           -0.001 ** (0.000)
  Females                               -0.108 *** (0.013)
  Blacks                                -0.080 *** (0.011)
  Hispanics                              0.022 (0.013)
  Married                                0.083 *** (0.012)
  Married females                       -0.104 *** (0.016)
  Education                              0.067 *** (0.002)

Place of residence
  Urban                                  0.035 *** (0.008)
  North Central                         -0.133 *** (0.015)
  South                                 -0.172 *** (0.014)
  West                                  -0.033 ** (0.016)

Industry
  Construction/mining                    0.259 *** (0.039)
  Manufacturing                          0.207 *** (0.036)
  Transportation, communications,        0.271 *** (0.039)
  and utilities
  Retail/wholesale trade                 0.018 (0.038)
  Finance, insurance,                    0.267 *** (0.040)
  and real estate
  Business services                      0.186 *** (0.039)
  Personal services                      0.060 (0.037)
  Professional services                  0.136 *** (0.037)
  Public administration                  0.060 (0.038)

Skill category
  Technical/sales workers               -0.028 ** (0.014)
  Clerical workers                      -0.148 *** (0.010)
  Service workers                       -0.177 *** (0.012)
  Operators/laborers                    -0.165 *** (0.013)
  Skilled laborers                      -0.116 *** (0.013)

Year dummy
  1996                                   0.050 *** (0.007)
  1998                                   0.103 *** (0.010)

N                                   16,919
Adjusted [R.sup.2]                  0.42

  * p < 0.1.

 ** p < 0.05.

*** p < 0.01.

The authors are more than usually indebted to two referees for their particularly helpful comments on earlier drafts of this paper and to our co-editor for her perseverance.

Received February 2005; accepted April 2006.

References

Abraham, Katherine G., and Susan K. Taylor. 1996. Firms' use of outside contractors: Theory and evidence. Journal of Labor Economics 14:394-424.

Addison, John T., and Christopher J. Surfield. 2006. The use of alternative work arrangements by the jobless: Evidence from the CAEAS/CPS. Journal of Labor Research 27:15-28.

Autor, David H. 2001. Why do temporary help firms provide free general skills training? Quarterly Journal of Economics 116:1409-49.

Autor, David H. 2003. Outsourcing at will: The contribution of unjust dismissal doctrine to the growth of employment outsourcing. Journal of Labor Economics 21:1-42.

Blackburn, McKinley L. 2007. Estimating wage differentials without logarithms. Labour Economics. In press.

Blank, Rebecca M. 1998. Contingent work in a changing labor market. In Generating jobs: How to increase demand for less-skilled workers, edited by Richard B. Freeman and Peter Gottschatk. New York: Russell Sage Foundation.

Booth, Alison L., Marco Francesconi, and Jeff Frank. 2002. Temporary jobs: Stepping stones or dead ends? Economic Journal 112:F189-213.

Castro, Janice. 1993. Disposable workers. Time 140:43-7.

Cohany, Sharon R. 1998. Workers in alternative employment arrangements: A second look. Monthly Labor Review 121:3-21.

Forde, Chris, and Gary Slater. 2005. Agency working in Britain: Character, consequences and regulation. British Journal of Industrial Relations 43:249-71.

Heinrich, Carolyn J., Peter R. Mueser, and Kenneth Troske. 2005. Welfare to temporary work: Implications for labor market outcomes. Review of Economics and Statistics 87:154-73.

Hipple, Steven, and Jay Stewart. 1996. Earnings and benefits of workers in alternative work arrangements. Monthly Labor Review 119:46-54.

Hirsch, Barry T. 2005. Why do part-time workers earn less? The role of worker and job skills. Industrial and Labor Relations Review 58:525-51.

Hirsch, Barry T., and Edward J. Shumacher. 2004. Match bias in wage gap estimates due to earnings imputation. Journal of Labor Economics 22:689-722.

Houseman, Susan N., Arne L. Kalleberg, and George A. Erickcek. 2003. The role of temporary agency employment in tight labor markets. Industrial and Labor Relations Review 57:105-27.

Houseman, Susan N., and Anne E. Polivka. 2000. The implications of flexible staffing arrangements for job security. In On the job: Is long-term employment a thing of the past? edited by David Neumark. New York: Russell Sage Foundation.

Madrian, Brigitte C., and Lars John Lefgren. 1999. A note on longitudinally matching Current Population Survey (CPS) respondents. NBER Technical Working Paper No. 247.

Manning, Willard G. 1998. The logged dependent variable, heteroskedasticity, and the retransformation problem. Journal of Health Economics 17:283-295.

Nollen, Stanley D. 1996. Negative aspects of temporary employment. Journal of Labor Research 27:567-82.

Polivka, Anne E. 1999. Using earnings data from the Current Population Survey after the redesign. Bureau of Labor Statistics Working Paper No. 306.

Polivka, Anne E., Sharon R. Cohany, and Steven Hipple. 2000. Definition, composition, and economic consequences of the nonstandard workforce. In Nonstandard work, edited by Francoise Carre et al. Ithaca, NY: Cornell University Press.

Rothstein, Donna S. 1996. Entry into and consequences of nonstandard work arrangements. Monthly Labor Review 119:76-83.

Segal, Lewis M., and Daniel G. Sullivan. 1997. The growth of temporary services work. Journal of Economic Perspectives 11:117-36.

Segal, Lewis M., and Daniel G. Sullivan. 1998. Wage differentials for temporary services work: Evidence from administrative data. Federal Reserve Bank of Chicago Working Paper WP-98-23.

(1) In what follows, we focus on supply-side wages necessary to attract/retain workers rather than the demand price to employers. Clearly, there is a wedge between the two, the price being paid to agency temporary firms by the client employer exceeding the wage received by the agency temps.

(2) Noting a large degree of skewness in earnings, particularly for contractors, Polivka, Cohany, and Hipple (2000) use hourly wages rather than the natural log of the hourly wage in their OLS regressions. Our imputations of the differential shown in the table are obtained by expressing the dollar estimate from the authors' OLS regressions as a percentage of mean reported regular worker earnings in 1995 or 1997. See also Appendix A.

(3) As noted by Segal and Sullivan (1998), one key advantage in using the unemployment insurance administrative data over the CPS is that the source of the information on industrial affiliation is the paycheck-issuing entity (for temporary workers, this would be the temporary help service agency). A concern with worker-reported data (as with the outgoing rotations of the CPS) is that agency temporary workers may cite the industrial affiliation of their client firm rather than that of their true employer--the temporary agency--in which case researchers will fail to identify temporary workers.

(4) In 2001, due to a CPS programming error, the survey was not administered to the outgoing rotation group. For the 2001 sample, we include only those regular workers who are identified in both the February CAEAS and the outgoing rotation group in March.

(5) Full OLS results for the pooled CAEAS sample are provided in Appendix C.

(6) Full OLS results for the NLSY79 sample are provided in Appendix D.

(7) Construction of the mathematical ability measure first required that we sum across the scores for the respondent's arithmetic reasoning, knowledge of mathematics, and numerical operations tests. In similar vein, practical ability was derived from adding the component scores for general science, auto/shop knowledge, mechanical comprehension, and electronics information. Only the coding measure involved no initial summation.

John T. Addison * and Christopher J. Surfield ([dagger])

* Department of Economics, Moore School of Business, University of South Carolina, 1705 College Street, Columbia, SC 29208, USA; E-mail [e-mail]; corresponding author.

([dagger]) Department of Economics, College of Business and Management, Saginaw Valley State University, 7400 Bay Road, University Center, MI 48710, USA; E-mail [e-mail].

Table 1. Selected U.S. Studies Examining the Effect of Alternate
Work Arrangements on Wages

Study                 Data                    Methodology

Hipple and            1995 Contingent         Cross-tabulations
  Stewart               and Alternative         alone
  (1996) (a)            Employment
                        Arrangement
                        Supplement
                        to the Current
                        Population
                        Survey
                        (CAEAS/CPS)
Rothstein             1994 National           Cross-tabulations
  (1996)                Longitudinal
                        Survey of Youth
                        (1979 Cohort)
Nollen (1996)         1984 Industry           Cross-tabulations
                        Wage Survey,
                        and 1994
                        Occupation
                        Compensation
                        Survey of
                        Temporary Help
                        Supply Services
Segal and             1983-1994 CPS           Cross-tabulations
  Sullivan (1997)       outgoing                plus cross-section
                        rotations               and fixed effects
                                                ordinary least
                                                squares (OLS)
                                                estimations
Blank (1998) (a)      1995 and 1996           Cross-tabulations
                        March Current
                        Population
                        Surveys
Cohany (1998) (a)     1997                    Cross-tabulations
                        CAEAS/CPS
Segal and             1984-1994               OLS estimation
  Sullivan (1998)       Washington              using cross-section
                        State                   data, plus worker
                        unemployment            fixed effects
                        insurance               estimation
                        administrative
                        data
Polivka,              1995 and 1997           OLS estimation
  Cohany, and           CAEAS/CPS
  Hipple (2000) (a)
Autor (2001)          1994 Occupation         OLS estimation
                        Compensation            using cross-section
                        Survey of               data, firm fixed
                        Temporary Help          effects estimation
                        Supply Services
Heinrich,             Administrative          Cross-tabulations
  Mueser, and           data for female         plus OLS earnings
  Troske (2005)         recipients of Aid       estimates, with
                        to Families with        selection
                        Dependent               argument for job
                        Children and            choice (derived
                        Temporary               from
                        Assistance for          a multinomial
                        Needy Families          logit) to control
                        in Missouri             for unobserved
                        (1993-1997) and         individual
                        North Carolina          heterogeneity
                        (1997)

Study                 Compensation            Work Form(s) Examined
                      Measure(s)

Hipple and            Median weekly           Agency temporary,
  Stewart               earnings                on-call, and
  (1996) (a)                                    contract workers;
                                                independent
                                                contractors; and
                                                regular/traditional
                                                workers
Rothstein             Average hourly          Agency and
  (1996)                earnings, and           direct-hire
                        average weekly          temporary workers,
                        hours                   contractors/
                                                consultants, and
                                                regular workers
Nollen (1996)         Average hourly          Agency temporary
                        wages                   employment and
                                                non-temporary
                                                (i.e., regular)
                                                employment
Segal and             Average hourly          Agency temporary
  Sullivan (1997)       wages                   employment and
                                                non-temporary
                                                (permanent)
                                                employment
Blank (1998) (a)      Hourly earnings         Agency temporary
                                                workers, regular
                                                workers
Cohany (1998) (a)     Median weekly           Agency temporary,
                        earnings                on-call, and
                                                contract workers;
                                                independent
                                                contractors; and
                                                regular/traditional
                                                workers
Segal and             Quarterly earnings      Agency temporary
  Sullivan (1998)       and quarterly           employment, and
                        hours worked            non-temporary
                        used in calculation     (regular)
                        of average hourly       employment
                        earnings
Polivka,              Hourly earnings         Agency temporary,
  Cohany, and                                   direct-hire
  Hipple (2000) (a)                             temporary, on-call,
                                                contract workers,
                                                independent
                                                contractors, and
                                                regular/traditional
                                                workers
Autor (2001)          Hourly wages;           Agency temporary
                        training in word        employment only
                        processing, data
                        entry, computer
                        programming
Heinrich,             Quarterly earnings      Welfare recipients in
  Mueser, and           (the current and        temporary help
  Troske (2005)         subsequent              service industry
                        earnings of             and other
                        welfare recipients      industries
                        in the temporary
                        help sector are
                        compared with
                        other employed
                        welfare recipients
                        in other
                        industries). The
                        study also
                        examines
                        employment and
                        welfare dynamics.

Study                 Findings

Hipple and            Median weekly earnings of agency
  Stewart               temporary workers ($290) and on-call
  (1996) (a)            workers ($386) are lower than those
                        of regular workers ($480). Contract
                        employment ($512) and independent
                        contracting ($518) carry a premium.
Rothstein             Atypical work is associated with lesser
  (1996)                experience: Both agency and direct-hire
                        temporaries spent a lower fraction of the
                        weeks prior to their job in employment
                        (46% and 48%, respectively) as
                        compared with regular workers (75%).
                        The wages of agency temporaries and
                        direct-hire temporaries are 35% and 25%
                        lower than regular workers, respectively.
                        Contractors have similar earnings to
                        regular full-time workers.
Nollen (1996)         In 1994, agency temporaries earned
                        $7.74 per hour, versus $11.94 for
                        non-temporaries. In both 1984 and 1994,
                        the differential attaching to temporary
                        work approximated 34%, some of which
                        is allied to the disproportionate number
                        of temporaries concentrated in
                        low-paying occupations, such as clerical
                        or laborer positions.
Segal and             Agency temporaries earned 78 cents on
  Sullivan (1997)       the permanent-worker dollar in 1993.
                        Cross-section OLS estimates controlling
                        for demographic, industrial, and
                        occupation differences are in the range
                        -8% to -14%. Fixed effects regressions
                        point to a negative differential of just
                        3%, suggesting that much of the wage
                        difference is attributable to unobserved
                        worker heterogeneity.
Blank (1998) (a)      On average, agency temporaries work
                        36 hours per week, as compared with
                        43 hours for regular workers, and
                        receive hourly wages that are 70% of
                        those paid to regular workers.
Cohany (1998) (a)     Findings similar to Hipple and Stewart
                        (1996) in row 1. Agency temporary
                        workers earn the least ($329 per week)
                        when compared with regular workers
                        ($510), and on-call work pays 85 cents
                        on the regular-employee dollar. Both
                        groups are younger and more likely to
                        be female or a minority member than
                        are regular workers. For their part,
                        contract workers and independent
                        contractors have median weekly
                        earnings of $619 and $587.
Segal and             The cross-section findings point to
  Sullivan (1998)       a statistically significant negative
                        earnings differential for agency
                        temporary work in excess of 30%.
                        But controlling for worker-specific
                        fixed effects reduces this to between
                        10% and 15%.
Polivka,              Cross-tabulations of hourly wages yield
  Cohany, and           results similar to Cohany (1998) in
  Hipple (2000) (a)     row 6. After taking into consideration
                        differences in worker characteristics,
                        the differential attached to agency
                        temporary (contracting) work is
                        estimated to be -5% (23%) in 1997.
Autor (2001)          Temporary workers offered training
                        were paid 2% to 3% lower wages
                        than those not given training. Firm
                        fixed effects results are of a negative
                        differential in excess of 5%.
Heinrich,             Welfare recipients in temporary jobs
  Mueser, and           (defined by sector) receive lower earnings
  Troske (2005)         than their counterparts in other jobs
                        initially, but after two years their
                        earnings are virtually identical to those
                        received in other jobs. This implied faster
                        earnings growth arises in part because
                        workers in the temporary help industry
                        are more likely to move into higher-
                        paying industries over time. Accordingly,
                        temporary workers have appreciably
                        better prospects than those who are not
                        holding jobs initially (while vis-a-vis
                        those who held jobs in other sectors they
                        are no more likely to be unemployed or
                        have appreciably higher rates of welfare
                        recidivism).

(a) Indicates that the study also examined health insurance benefit
access/coverage (together with eligibility for employer-related
pensions in the case of the studies in rows 1, 5, and 6).

Table 2. Mean Hourly Wage Rates by Employment Arrangement,
CAEAS/CPS Data

                              1995            1997

Regular workers           13.47 (7.71)    14.07 (7.59)
Screened workers          13.49 (6.99)    14.13 (6.52)
Agency temporaries         9.18 (6.81)    10.98 (8.67)
Direct-hire temporaries   12.64 (8.69)    13.81 (10.23)
On-call workers           12.61 (12.11)   12.31 (10.66)
Contract workers          14.23 (10.27)   15.76 (9.35)
Independent contractors   16.29 (13.21)   18.18 (15.26)
n                            21,885          21,177

                              1999            2001

Regular workers           15.62 (9.32)    17.22 (10.32)
Screened workers          16.37 (9.31)    17.34 (10.44)
Agency temporaries        12.00 (13.29)   14.87 (11.76)
Direct-hire temporaries   14.86 (11.23)   16.35 (11.86)
On-call workers           14.42 (13.58)   15.61 (13.96)
Contract workers          17.42 (9.13)    19.78 (12.30)
Independent contractors   19.74 (16.78)   21.12 (17.74)
n                            21,628           8839

Standard deviations in parentheses.

Table 3. Mean Hourly Wage Rates by Employment Arrangement, NLSY79 Data

                              1994            1996            1998

Regular workers           13.57 (9.43)    14.94 (11.22)   16.34 (11.80)
Screened workers          13.01 (7.56)    15.29 (9.73)    15.67 (9.92)
Agency temporaries         7.70 (4.47)     9.53 (5.17)     9.34 (5.21)
Direct-hire temporaries    9.88 (4.31)     9.67 (5.64)     9.79 (4.64)
Contractors/consultants   16.49 (20.38)   21.29 (19.03)   23.53 (23.12)
Other work types          19.79 (15.85)   19.92 (17.82)   17.78 (6.91)
N                             5551            5702            5666

Standard deviations in parentheses.

Table 4. Labor Market Experience by Employment Arrangement

                            Regular        Screened        Agency
                            Workers        Workers      Temporaries

1994 NLSY79 data
  Experience (in years)   13.03 (3.33)   12.56 (3.24)   10.13 (4.19)
  Tenure (in years)        5.24 (4.52)    5.01 (4.13)    1.25 (1.77)
  Jobs (standardized       0.78 (0.56)    0.85 (0.57)    1.67 (1.46)
    number of jobs)
N                             4910           427             51

                          Direct-Hire    Contractors/    Other Work
                          Temporaries    Consultants       Types

1994 NLSY79 data
  Experience (in years)   10.39 (3.76)   12.69 (3.77)   13.11 (3.58)
  Tenure (in years)        1.36 (2.57)    3.96 (3.73)    2.11 (2.36)
  Jobs (standardized       1.49 (l.05)    0.97 (0.55)    1.08 (0.54)
    number of jobs)
N                              47             86             30

                            Regular        Screened        Agency
                            Workers        Workers      Temporaries

Pooled CASAS/CPS data
  Tenure (in years)       8.01 (8.01)    6.96 (6.63)    1.53 (2.40)
  N                          53,995          2018           1037

                          Direct-Hire      On-Call
                          Temporaries      Workers

Pooled CASAS/CPS data
  Tenure (in years)       5.72 (7.53)    4.86 (6.00)
  N                           4954           613

                            Contract     Independent
                            Workers      Contractors

Pooled CASAS/CPS data
  Tenure (in years)       4.47 (6.35)    9.59 (8.57)
  N                           1868           9125

Standard deviations in parentheses.

Table 5. OLS Regression Estimates of Atypical Worker Wage
Differentials, Pooled CAEAS/CPS Data (Dependent Variable:
Log Hourly Wage)

                             (1)

Screened
  workers             0.007 (0.009)
Agency
  temporaries        -0.185 *** (0.015)
Direct-hire
  temporaries        -0.105 *** (0.007)
On-call workers      -0.089 *** (0.014)
Contract
  workers             0.083 *** (0.020)
Independent
  contractors         0.001 (0.008)
Tenure
[Tenure.sup.2]/
  100
Allocated?                    Y
n                          73,610
Adjusted
  [R.sup.2]                 0.34

Using the results obtained from the full samples, F-tests
were conducted to determine the following restrictions,
with test statistic [rho] and p value (in parentheses)
reported below:

                                              Using Allocated Workers

[beta]AT =
  [beta]DT                                    [rho] = 22.98 (p = 0.000)
[beta]CW =
  [beta]IC                                    [rho] = 14.83 (p = 0.000)
[beta]AT =
  [beta]DT =
  [beta]OC =
  [beta]CW =
  [beta]IC                                    [rho] = 55.95 (p = 0.000)

                             (2)                         (3)

Screened
  workers             0.012 (0.008)               0.008 (0.011)
Agency
  temporaries        -0.109 *** (0.015)          -0.175 *** (0.019)
Direct-hire
  temporaries        -0.072 *** (0.007)          -0.114 *** (0.009)
On-call workers      -0.052 *** (0.013)          -0.104 *** (0.017)
Contract
  workers             0.101 *** (0.020)           0.100 *** (0.027)
Independent
  contractors        -0.013 ** (0.008)            0.009 (0.010)
Tenure                0.024 *** (0.001)
[Tenure.sup.2]/
  100                -0.041 *** (0.002)
Allocated?                    Y                           N
n                          73,610                      43,465
Adjusted
  [R.sup.2]                 0.36                        0.34

Using the results obtained from the full samples, F-tests were
conducted to determine the following restrictions, with test
statistic [rho] and p value (in parentheses) reported below:

[beta]AT =
  [beta]DT
[beta]CW =
  [beta]IC
[beta]AT =
  [beta]DT =
  [beta]OC =
  [beta]CW =
  [beta]IC

                                                         (4)

Screened
  workers                                         0.014 (0.010)
Agency
  temporaries                                    -0.090 *** (0.019)
Direct-hire
  temporaries                                    -0.081 *** (0.009)
On-call workers                                  -0.059 *** (0.017)
Contract
  workers                                         0.115 *** (0.027)
Independent
  contractors                                    -0.011 (0.010)
Tenure                                            0.026 *** (0.001)
[Tenure.sup.2]/
  100                                            -0.042 *** (0.003)
Allocated?                                                N
n                                                      43,465
Adjusted
  [R.sup.2]                                             0.38

Using the results obtained from the full samples, F-tests were
conducted to determine the following restrictions, with test
statistic [rho] and p value (in parentheses) reported below:

                  Excluding Allocated Workers

[beta]AT =
  [beta]DT        [rho] = 8.85 (p = 0.003)
[beta]CW =
  [beta]IC        [rho] = 10.62 (p = 0.001)
[beta]AT =
  [beta]DT =
  [beta]OC =
  [beta]CW =
  [beta]IC        [rho] = 41.67 (p = 0.000)

Huber-White standard errors to correct for heteroskedasticity
are given in parentheses. The differences in sample size across
the allocated and nonallocated samples are attributable to the
loss of the 1995 cross section (see text). Additional controls
are age (and [age.sup.2]), gender and ethnicity, a dummy variable
equal to one if married (zero otherwise), an interaction term between
gender (being female) and marital status, six educational dummies
(the omitted category is no high school diploma), a dummy variable
equal to one if residing in an urban area (zero otherwise), four
regional dummies (the omitted category is residing in the Northeast),
10 industry dummies (the omitted category is
agriculture/fishing/forestry), and six occupational
dummies (the omitted category is manager).

* p < 0.1.

** p < 0.05.

*** p < 0.01.

Table 6. Corrected Ordinary Least Squares Regression Estimates
of Atypical Worker Wage Differentials, Pooled CAEAS/CPS Data

                                 (1)                  (2)

Screened workers          -0.004 (0.006)        0.000 (0.006)
Agency temporaries        -0.148 *** (0.008)   -0.084 *** (0.009)
Direct-hire temporaries   -0.057 *** (0.004)   -0.030 *** (0.004)
On-call workers           -0.011 * (0.007)      0.021 *** (0.007)
Contract workers           0.115 *** (0.013)    0.137 *** (0.014)
Independent contractors    0.162 *** (0.004)    0.147 *** (0.004)

                                 (3)                  (4)

Screened workers           0.001 (0.008)        0.005 (0.008)
Agency temporaries        -0.132 *** (0.010)   -0.058 *** (0.011)
Direct-hire temporaries   -0.057 *** (0.005)   -0.030 *** (0.005)
On-call workers           -0.022 *** (0.008)    0.017 * (0.009)
Contract workers           0.143 *** (0.018)    0.168 *** (0.018)
Independent contractors    0.183 *** (0.005)    0.166 *** (0.005)

Columns 1 through 4 correspond to those in Table 5. See also the
notes to Table 5.

* p < 0.1.

** p < 0.05.

*** p < 0.01.

Table 7. Random Effects Ordinary Least Squares Estimates of
Atypical Worker Wage Differentials, NLSY79 Data

                                 (1)                  (2)

Screened workers                0.011           0.035 *** (0.002)
Agency temporaries        -0.113 *** (0.036)   -0.300 *** (0.006)
Direct-hire temporaries    0.002 (0.038)        0.013 (0.012)
Contractors/consultants    0.075 * (0.043)      0.304 *** (0.010)
Other work types           0.117 ** (0.049)     0.504 *** (0.020)
Experience                 0.010 * (0.005)      0.010 * (0.005)
Experience (2)             0.001 *** (0.000)    0.001 *** (0.000)
Tenure                     0.028 *** (0.002)    0.028 *** (0.002)
Tenure (2)                -0.001 *** (0.000)   -0.001 *** (0.000)
Jobs (standardized)       -0.032 *** (0.009)   -0.032 *** (0.009)
Coding ability
Math ability
Practical ability
Verbal ability
Adjusted [R.sup.2]               0.42
N                               16,919               16,919

                                 (3)                  (4)

Screened workers          0.024 *** (0.003)   -0.016 *** (0.004)
Agency temporaries       -0.319 *** (0.007)   -0.451 *** (0.009)
Direct-hire temporaries   0.032 *** (0.012)    0.064 *** (0.017)
Contractors/consultants   0.419 *** (0.013)    0.328 *** (0.013)
Other work types          0.635 *** (0.025)    0.516 *** (0.030)
Experience                0.012 *** (0.005)    0.024 *** (0.008)
Experience (2)            0.001 *** (0.000)    0.001 (0.000)
Tenure                    0.027 *** (0.002)    0.022 *** (0.003)
Tenure (2)               -0.001 *** (0.000)   -0.001 *** (0.000)
Jobs (standardized)      -0.026 *** (0.009)   -0.034 ** (0.015)
Coding ability            0.014 ** (0.006)     0.022 *** (0.009)
Math ability              0.058 *** (0.009)    0.059 *** (0.012)
Practical ability         0.010 (0.010)        0.002 (0.013)
Verbal ability            0.012 (0.008)        0.015 (0.011)
Adjusted [R.sup.2]               0.43                 0.40
N                               16,106                8570

                                 (5)

Screened workers           0.075 *** (0.004)
Agency temporaries        -0.133 *** (0.011)
Direct-hire temporaries    0.011 * (0.006)
Contractors/consultants    0.626 *** (0.038)
Other work types           0.888 *** (0.066)
Experience                 0.008 *** (0.001)
Experience (2)             0.001 *** (0.000)
Tenure                     0.033 *** (0.003)
Tenure (2)                -0.001 *** (0.000)
Jobs (standardized)       -0.017 (0.012)
Coding ability             0.009 (0.009)
Math ability               0.058 *** (0.012)
Practical ability          0.004 (0.017)
Verbal ability             0.015 (0.012)
Adjusted [R.sup.2]               0.44
N                                7536

F-tests were conducted to determine the following
restrictions, with test statistic p and p value
(in parentheses) reported below:

                            Full Sample,         Full Sample,
                           Without Ability        With Ability
                               Proxies              Proxies

[beta]AT = [beta]DT          [rho] = 5.30         [rho] = 6.57
                             (p = 0.021)          (p = 0.010)

[beta]AT = [beta]DT =       [rho] = 18.93        [rho] = 23.99
  [beta]C/C =                 (p = 0.00)           (p = 0.00)
  [beta]OTHER

Huber-White standard errors to correct for heteroskedasticity
in parentheses. Columns 1 though 3 are for all workers, while
4 and 5 pertain to males and females, respectively. Additional
controls are age (and age 2), gender and ethnicity, a dummy
variable equal to one if married (zero otherwise), an interaction
term between gender (being female) and marital status, education
(in years), a dummy variable equal to one if residing in an urban
area (zero otherwise), four regional dummies (the omitted category
is residing in the Northeast), 10 industry dummies (the omitted
category is agriculture/fishing/forestry), and six occupational
dummies (the omitted category is manager).

* p < 0.1.

** p < 0.05.

*** p < 0.01.

Table 8. Corrected Fixed Effects Ordinary Least Squares Estimates
of Atypical Worker Wage Differentials, NLSY79 Data

                             Full Sample             Males

Screened workers          0.045 *** (0.002)    0.013 *** (0.003)
Agency temporaries       -0.165 *** (0.005)   -0.335 *** (0.009)
Direct-hire temporaries   0.118 *** (0.009)    0.153 *** (0.017)
Contractors/consultants   0.196 *** (0.008)    0.311 *** (0.016)
Other work types          0.390 *** (0.019)    0.149 *** (0.018)
Experience                0.087 *** (0.022)    0.144 *** (0.030)
Experience (2)            0.001 (0.000)       -0.001 * (0.000)
Tenure                    0.017 *** (0.003)    0.012 *** (0.004)
Tenure (2)               -0.001 *** (0.000)   -0.001 ** (0.000)
Jobs (standardized)      -0.018 (0.022)        0.002 (0.025)
N                              16,919                9084
Adjusted [R.sup.2]              0.27                 0.16

F-tests were conducted to determine the following
restrictions, with test statistic p and p-value
(in parentheses) reported below:

                             Full Sample           Males Only

[beta]AT = [beta]DT         [rho] = 2.06          [rho] = 4.53
                              (p=0.151)            (p=0.033)

RAT = PDT =                 [rho] = 1.78         [rho] = 3.86
  [beta]C/C =                (p = 0.149)          (p = 0.009)
  [beta]OTHER

                               Females

Screened workers           0.087 *** (0.002)
Agency temporaries         0.151 *** (0.014)
Direct-hire temporaries    0.113 *** (0.011)
Contractors/consultants   -0.032 *** (0.011)
Other work types           1.223 *** (0.107)
Experience                 0.038 (0.030)
Experience (2)             0.000 (0.000)
Tenure                     0.023 *** (0.004)
Tenure (2)                -0.001 *** (0.000)
Jobs (standardized)       -0.033 (0.036)
N                                7835
Adjusted [R.sup.2]               0.28

F-tests were conducted to determine the following
restrictions, with test statistic p and p-value
(in parentheses) reported below:

                             Females Only

[beta]AT = [beta]DT         [rho] = 0.01
                             (p = 0.915)

RAT = PDT =                  [rho] = 1.26
  [beta]C/C =                (p = 0.287)
  [beta]OTHER

Additional controls are age (and [age.sup.2]), a
dummy variable equal to one if married (zero otherwise),
an interaction term between gender (being female) and
marital status, education (in years), a dummy variable
equal to one if residing in an urban area (zero otherwise),
four regional dummies (omitted category is residing in the
Northeast), 10 industrial dummies (omitted category is
working in agriculture/fishing/forestry), and six
occupational dummies (omitted category is employment
as a manager).

* p < 0.1.

** p < 0.05.

*** p < 0.01.