The Program and Treatment Effect of Summer Jobs on Girls’ Post-Schooling Incomes

Abstract

Background:

Public programs offering summer jobs to smooth the transition from school to work is commonplace. However, the empirical support for summer jobs is limited. This article exploits the availability of registered individual information and random allocation to summer jobs to provide empirical evidence on this issue.

Objectives:

To identify the effect of summer job programs on the post-schooling incomes of the intended participants. Also to identify the effect of sophomore girls’ high school work experience on their post-schooling incomes.

Research Design:

In this article, 1,447 sophomore girls from 1997 to 2003 are followed 5–12 years after graduation. They all applied to Falun municipality’s (Sweden) summer job program, and about 25% of them were randomly allotted a job. The random allocation to a summer job is used to identify the causal effect of sophomore girls’ high school income on their post-schooling incomes.

Subjects:

All the 1,447 sophomore girls who applied to Falun municipality’s summer job program during 1997–2003.

Measures:

Annual post-schooling income is used as an outcome measure. The work experience of girls in high school is also measured in terms of total income while in high school.

Results:

The program led to a substantially larger accumulation of income during high school as well as 19% higher post-schooling incomes. The high school income led to a post-schooling income elasticity of 0.37 which is, however, potentially heterogeneous with regard to academic ability.

Conclusions:

Both the program effect and the causal effect of high school income on post-schooling incomes were substantial and statistically significant.

Keywords

experimental data work experience work while in school selection bias

Introduction

Is early contact with the labor market, through for instance summer jobs, economically beneficial for high school students in the long run? The answer carries important implications for labor policy makers worldwide in deciding whether to aid in smoothing the transition from high school to work. Indeed, public programs promoting early labor market contacts for high school students are commonplace and they have repeatedly been advocated by the Organization for Economic Cooperation and Development (OECD).¹ In Sweden, the government commenced subsidizing summer jobs in 1995. The government administers the subsidies via the Swedish municipalities. The municipality offers, within its organization, summer jobs to high school students.

It is widely believed that the summer job experience is beneficial to high school students and their future labor market outcomes. One argument in favor of this is that the initial contacts with employers lead to more and better subsequent contacts (see, e.g., Rosenbaum, DeLuca, Miller, & Roy, 1999; Alon, Donahoe, & Tienda, 2001; Carling & Larsson, 2005; Carr, Wright, & Brody, 1996; Geel & Backes-Gellner, 2012;² Hensvik & Nordström-Skans, 2013; Häkkinen, 2006; Ruhm, 1997). Counterarguments are however also available in the literature (see the review in Ruhm, 1997; see also Lee & Orazem, 2010; Staff, Schulenberg, & Bachman, 2010; Weller, Cooper, Basen-Engquist, Kelder, & Tortolero, 2003 and references therein).

In spite of this belief, the empirical support for summer job programs is limited (Leos-Urbel, 2012; Wang, Carling, & Nääs, 2006). As for the broader research topic on “working while in high school,” the focus has been on school attendance, school grades, and disposable income emphasizing the potential negative consequences of work experiences (see the review in Ruhm, 1997).

It is a difficult task to empirically determine the effect of a summer job experience. First, there are few data sets suitable for this purpose since information about summer jobs and their holders are rarely documented. Second, there are methodological challenges where selection bias is the foremost (see Hotz, Xu, Tienda, & Ahituv, 2002; Monahan, Lee, & Steinberg, 2011). A summer job is presumably the result of an active job-searching process, and any correlation between a summer job experience and later outcomes may be due to unobserved individual abilities rather than being a causal relationship.

In this article, we estimate the effect of high school girls’ summer job program participation (program effect) and the effect of their work experience (treatment effect) on post-schooling incomes. We study 1,447 sophomores in Falun (a medium-sized town in central Sweden) and exploit the random allocation of summer job among them to overcome the problem of selection bias.

This article is organized as follows: In the second section, we describe the data and the program. In the third section, we estimate the program effect on the post-schooling incomes. In the fourth section, we examine the accumulated work experience during high school and estimate the causal effect of accumulated work experience on post-schooling incomes. Finally, the fifth section concludes with a discussion of our findings and how they relate to current literature on the topic.

Data and the Summer Job Program

The data come from two sources. One source is Statistics Sweden (SCB). SCB has provided data about all individuals who attended high school in the country. The data contain background variables, such as gender, parental socioeconomic status, and school grades, as well as outcome data in the form of annual incomes³ up to a maximum of 31 years of age. The other source is the Falun municipality that has been running a summer job program since 1995 for 16- to 18-year-old students. We received data pertaining to all applicants during the years 1997–2003. The data contain the name and contact address as well as civic registration number. The two data sources were merged by SCB by using the civic registration number as the key variable. The summer jobs were allocated randomly among the applicants for reasons of fairness as claimed by the officials. The allocation was in the form of an electronic lottery carried out by an official using an Excel spreadsheet. The data also contain information on who was randomly offered a summer job as well as whether the offer was accepted.⁴ All the data analysis was conducted by using the SAS software via SCB’s remote data access and analysis facility named Micro data Online Access (www.scb.se/mona).

The Swedish municipalities collect taxes to fund and operate elderly care centers, recreational areas, schools, social aid as well as a local street network. The summer jobs in the program consisted, however, of unskilled job tasks, such as assisting in elderly care, gardening, and cleaning.

Regarding the students targeted by the program, it is useful to know that the Swedish high school runs for 3 years, and the students typically commence high school at the age of 16. We will refer to sophomore, junior, and senior to denote a high school student’s first, second, and third year in high school. Although high schooling is voluntary, almost all teenagers are enrolled (OECD, 2014). High school programs are divided into two categories, consisting of academic programs targeting university studies and vocational programs aiming at a direct transition to the labor market. A reform prior to the study period rendered students of the vocational programs eligible for university, and about one third of high school students are later enrolled in a university (UKÄ, 2013; OECD, 2014).

The fall semester of the high school starts in the mid of August and the spring semester ends in mid-June. Consequently, the students are available for summer jobs for about a 9-week period between June and August.⁵ We restrict our focus to high school sophomore girls during the years 1997–2003.⁶ It was possible to apply for summer jobs in each high school year and in each of the years the applicants were randomized. However, the reason to focus on sophomores is methodological. We thereby avoid the risk of introducing bias when addressing the task of understanding and modeling the process of repeated applications to the program in the course of high school.

The number of summer jobs offered within the program was predetermined each year by the municipality. The duration of the summer jobs was normally 3 weeks and the municipality paid about Swedish Crown (SEK) 63 (approximately US$10) per hour, or SEK 2,500 per week, at the price level of 2009. It is worth noting that the state or the government exercises no influence on minimum wage in Sweden. Minimum wage is the result of collective bargaining between the union and the employers’ association within the sector. The minimum wages for a person below 17 years of age was SEK 66 in the manufacturing sector, and for nonprofit organizations, real estate, hotels and restaurants, municipality, and retail it was SEK 59, 54, 61, 53, and 64, respectively. These are the sectors where unskilled, young people are mostly in demand. Given the lack of previous work experience, it is hard to imagine a substantial variation in wages of the sophomores, unless the supply is scarce relative to the demand, in which case the union ought to be able to raise the minimum wage. Thus, the wage of SEK 63 paid to the girls in the program is clearly a competitive wage relative to alternative summer jobs. In the following, all the hourly wages and weeks of work experiences are derived by using SEK 2,500 as equivalent to the wage for 1 week’s fulltime (40 hr) work.

Altogether there were 3,220 sophomore girls in Falun in the study period, and 1,447 of these applied for a summer job via the program, while the remaining 1,773 did not apply.⁷ Figure 1 depicts the girls divided into subgroups of applicants and nonapplicants in the course of their high school studies. For instance, the 1,447 girls applying in their sophomore year are divided into juniors as follows: 566 re-applying, 572 not re-applying, and 309 for whom data were missing. In the senior year, the 566 re-applicants are divided into 18 re-re-applicants, 450 not applying, and 98 for whom data were missing in their senior year only. The missing data arise primarily because sophomores in 2003 (and 2002) were not observed as juniors (and seniors) due to the data restricted to 1997–2003. A secondary reason is that some girls moved from Falun. Note that we still have the girls’ post-schooling incomes (and other variables) retrieved from the second data source.

Figure 1.

Sophomore girls in the years 1997–2003 and the application to the summer job program. Applicants (A), nonapplicant (NA), offered (O), and missing (M) during the 3 years of high school.

The first point to notice in Figure 1 is that the applications in the senior year were very scarce. There were 44 applications submitted by seniors that make up only some 2% of all the 2,340 submitted applications. The regulations of the labor market are much less restrictive for seniors than younger students, and consequently the seniors stand a good chance of finding a summer job (at the municipality or elsewhere) without using the program as a mean. The 1,773 nonapplicants were rather unlikely to apply later in high school, only 283 of them applied in their junior year. However, applicants were likely to re-apply as juniors. Thus, the program contained mainly girls who applied as sophomores. This is another reason to focus the analysis on the 1,447 sophomores since they are those most affected by the existence of the program.

Table 1 shows descriptive statistics of the sophomores. Two important variables are also shown, namely school grade of compulsory schooling (final grade at the age of 15 years), and parents’ income calculated as father’s income from work plus mother’s income from work in SEK 100,000. The parental income was taken for the year the girl was a sophomore, since the income mostly varies between families rather than overtime. The two variables presumably correlate with the propensity of applying for the program, the chance of obtaining a summer job, and the work experience during high school as well as the post-schooling incomes (cf. Carr et al., 1996).

Table 1.

Descriptive Statistics of the Girls in the Sophomore Year in the Municipality of Falun in the Years 1997–2003.

Variable	Nonapplicants	Applicants
Variable	Nonapplicants	All	Offered	Not offered
Offered (%)	—	25.0	—	—
Grades^a	−0.1 (1.0)	0.1 (0.9)	0.1 (1.0)	0.2 (0.9)
Parental income^b	3.9 (2.4)	4.1 (2.1)	3.8 (2.1)	4.2 (2.1)
Summer jobs (%)^c	54.2	66.4	94.8	57.0
Income as sophomore^d	9.2 (14.9)	8.5 (10.1)	10.0 (7.6)	8.0 (11.7)
Birth cohort (%)
1980	17.8	7.7	5.8	8.3
1981	13.6	12.4	11.0	12.9
1982	12.2	15.0	16.9	14.4
1983	9.5	13.5	16.3	12.6
1984	10.8	15.4	19.9	13.9
1985	11.4	14.6	15.5	14.3
1986	10.8	11.7	10.2	12.3
1987	13.9	9.6	4.4	11.3
No	1,773	1,447	362	1,085

Note. For continuous variables, means are given and standard deviations are in parentheses.

^aOriginal grades from compulsory school are in different scales before and after 1998 so they were transformed to Z-scores to enable a comparison over time.

^bParents’ income when the student was in the sophomore year (SEK 100,000).

^cIncome exceeding SEK 2,500 (1 week of full-time work) is required.

^dSEK 1,000.

A number of points are worth noting in Table 1. First, there seems to be no major difference between applicants and nonapplicants, or those who were offered a summer job (offered) and those who were not offered a summer job (nonoffered) with regard to background variables. Second, the proportion of sophomores with some experience of a summer job was rather high, about 55% for nonapplicants and nonoffered applicants and (unsurprisingly) almost 100% for offered. The high uptake of summer jobs among the offered girls also resulted in a (25%) higher average income compared with nonoffered girls. Thirdly, over the years 1997–2003, a quarter of the applicants were offered a summer job, but there were variations between cohorts and years: the number of summer jobs (for both boys and girls) each year, starting with 1997, was 84, 342, 240, 158, 269, 156, and 176 in 2003. The variation in the number of offers per year points at a need for controlling for cohort effects in the analysis.

Table 2 shows the estimated logistic models for the probability of applying as a sophomore and the probability of applicants to re-apply as juniors. Grades and parental income are statistically insignificant, or their effect is modest in magnitude. The propensity of applying to the program varied between years, but not between students born in different quarters of the year. Moreover, greater work experience lowered the propensity of re-applying to the program. This is possibly because the student had acquired a greater network of potential employers. If a girl was offered a job as a sophomore, then she was more likely to re-apply as a junior, which might indicate a desire to return to the same work place. Table 2 also shows the estimated logistic model for the probability of being offered a summer job for the sophomores who applied to the program. If the officials executed the lottery properly, we should expect none of the background variables to influence the probability of being offered a job. Indeed, none of the variables are statistically significant.

Table 2.

Estimated Logistic Models for the Probabilities of Application, Offer, and Re-application.

	All Sophomores	Sophomores Who Applied
Variables	Prob. of Applying	Prob. of offer	Prob. of re-applying as a junior
Intercept	−0.46 (0.24)	−1.09* (0.56)	−0.25 (0.40)
Grades (in Z-score)	0.23* (0.11)	−0.04 (0.25)	0.11 (0.18)
Parental income (in SEK 100,000)	−0.01 (0.01)	−0.02 (0.02)	0.02* (0.01)
Income as sophomore (in SEK 1,000)	—	—	−0.07* (0.01)
Offered as sophomore	—	—	0.31* (0.15)
Birth quarter dummies	Yes	Yes	Yes*
Cohort dummies	Yes*	Yes	Yes*
Interacted with grades	Yes	Yes	Yes
Interacted with parental income	Yes	Yes	Yes
Number of observations	3,220	1,447	1,138

Note. Standard errors in parentheses. Number of observations differ between Columns 3 and 4 because some sophomores were not observed in their junior year (see Figure 1).

*Indicates significant at 5% level.

Program Effect of Being Offered a Summer Job as a Sophomore

Our main interest is not about the work experience or the income in the sophomore year, but if the summer job leads to a stronger future position on the labor market, manifested by for instance higher post-schooling incomes. However, prior to examining this issue, we will take a look at how the income from work accumulates during the high school. Table 3 shows the girls’ evolution of income while in high school.

Table 3.

The Accumulation of Work Experience While in High School (Income in SEK 1,000).

		Applicants
	Nonapplicants	Offered	Nonoffered	Difference (%)
Sophomore
Mean	9.2	10.0	8.0	25
Q1	0.0	5.4	0.0	–
Median	3.8	7.4	4.5	64
Q3	12.4	12.4	11.8	5
Junior
Mean	23.0	29.4	24.2	21
Q1	3.1	12.9	3.2	303
Median	14.5	25.4	17.8	43
Q3	32.1	39.2	34.1	15
Senior
Mean	57.9	78.9	65.0	21
Q1	13.3	38.4	29.0	32
Median	44.5	72.3	54.9	32
Q3	85.7	111.8	91.3	22
Number of observations	1,773	362	1,085	362/1,085

The first thing to notice is that most of the work experience came in the later part of schooling, but that the majority of the girls started to acquire some experience already at the beginning. The income in high school was similar for the nonapplicant and the nonoffered groups, whereas the offered group consistently had the highest income. The strongly statistically significant difference in mean income indicates that the girls who were offered jobs had some 20–25% higher incomes than those who were not offered jobs. The difference of 3 weeks in the sophomore year was further accentuated by the time of graduation when the difference in work experience had reached 6 weeks (32 vs. 26 weeks). Granovetter (cited in Rosenbaum et al., 1999, p. 180) writes “…early contacts may have a greater impact on later jobs than on early jobs. This effect may occur because of the accumulation of advantages that come from the initial contacts, since good initial contacts lead to more and better subsequent contacts.” Granovetter’s idea of early contact gradually accumulating to a greater advantage in the course of time seems applicable for these students.⁸ In conclusion, the program appears to have induced a persistent increase in work experience of some 25% during high school.

So what happened after graduation to the girls that were in the program? The summer job opportunity was, and still is, open to every high school student in Falun, and it is therefore reasonable to focus on the average program effect of being offered a summer job in the sophomore year. Because the summer job allocation is made via a lottery, we may estimate the program effect as:

Δ_{t} = E [Y_{t}^{O}] - E [Y_{t}^{\overset{ˉ}{O}}]

where $E [Y_{t}]$ is the expected post-schooling income (Y) after t years from graduation, either in the case of having received the offer of a summer job ( $O$ ) or not ( $\overset{ˉ}{O}$ ). We estimate the program effect on log income,⁹ meaning that the estimated program effect is the relative difference in post-schooling income. This is a common practice in modeling income (see, e.g., Angrist & Kruger, 1991; Hotz et al., 2002) and it is motivated by the positive skewness of the distribution of post-schooling income, implying that the mean poorly represents the center of the distribution.

Figure 2 shows the trend of average annual income (in log) up to 12 years after graduation for offered, nonoffered, and nonapplicant,¹⁰ classified according to their application/offer status in sophomore year. The visual impression of Figure 2, that the summer job program affected the girls positively, is supported by regression modeling and formal statistical testing. We regressed all post-schooling annual incomes (in log) after graduation on an indicator variable for girls who were offered summer jobs as opposed to nonoffered, including a variable that represents years after graduation. In the regression analysis, we gave due account to having repeated measures of post-schooling incomes of the applicants in line with Liang and Zeger (1986) and Littell, Milliken, Stroup, and Wolfinger (1996; more details in the fourth section). The estimated program effect on (log-) post-schooling income is 0.19 (standard error of 0.10) with a p-value of 0.03 for a one-sided test. Thus, it seems that the summer job program induced some 25% increase in work experience while in high school. In addition, it seems that the girls who were offered jobs had an increase in post-schooling income of 19% (approximately) when compared with those girls who were not offered jobs. Hence, the program appears to have been both of a short-term and of a long-term benefit.

Figure 2.

Trend of average annual (log) post-schooling income for offered, nonoffered, and nonapplicant sophomores.

Estimation of the Treatment Effect

Early contact with the labor market supposedly paves the student’s way to a better working career. The indicator variable offered or not is a poor measure of this contact. Presumably, the post-schooling effect is more related to the quantity of the work experience acquired in high school, as discussed by Carr, Wright, and Brody (1996) and Ruhm (1997), who used hours worked per week as a measure. The municipality offered 3 weeks of work experience of 40 hr a week, but the girls who were offered jobs received on average 4 weeks of work experience as sophomores. Further, the sophomores who were not offered jobs nonetheless acquired work experience (about a half of them had a summer job, cf. Tables 1 and 3). Consequently, offered and nonoffered girls differ in the amount of contact they had with the labor market as sophomores.

In fact, a random allocation to the program cannot generate a positive effect per se. A positive effect may arise only if the program leads to a summer job and the student gains valuable experience that helps the working career. This fact indicates the usefulness of random program allocation as an instrument to identify the causal effect of work experience while in high school (Angrist, Imbens, & Rubin, 1996).

In this section, the treatment variable (or dose) is the total work experience measured as the accumulated income while in high school. We accumulated the girls’ incomes up to the first semester in the senior year. We did so because we cannot separate the annual income between the second semester in the senior year and the fall after graduation. Table 3 showed however that the relative difference in mean income between those who were offered high school summer jobs and those who were not offered was similar over the high school years rendering this choice inconsequential. Given 43% more work experience for the offered (median value in Table 3) and an estimated increase in post-schooling income of 19%, we expect to find an elasticity of about 0.5.

We first estimate a model including all the 3,220 sophomores and using their accumulated income in high school as the treatment. The model is

ln (Y_{i t}) = α + γ ln ({D o s e}_{i}) + δ t_{i} + β_{1} x_{1} + \dots + β_{k} x_{k} + ϵ_{i t},

where t_i is the years after graduation for the ith girl and ${D o s e}_{i} = \sum_{t = - 2}^{- 1} l n (Y_{i t})$ is her accumulated income while in high school. To account for the correlation in the annual post-schooling incomes of each girl, we assume a compound symmetric correlation structure (Liang & Zeger, 1986) in Model (2). That is, with $V a r (ϵ_{i, t} | x_{i, t}^{T}) = σ^{2}$ the covariance is $C o v (ϵ_{i, t}, ϵ_{i, t^{'}} | x_{i, t}^{T}, x_{i, t^{'}}^{T}) = σ^{2} ρ$ for $\forall t \neq t^{'}$ where $x_{i, t}^{T} = (1, l n {(D o s e)}_{i}, t_{i}, x_{1}, \dots, x_{k})$ . Table 4 (left results column) shows the results. The (statistically significant) elasticity is estimated as 0.34. This estimation procedure corresponds to the conventional observational analysis of the effect of work experience while in high school as conducted before by, for example, Carr et al. (1996) and Ruhm (1997).

Table 4.

Estimates of Model (2) for the Sophomores.

Variable	All Girls (GEE Estimate)	Only Girls Who Applied (IV Estimate)
Intercept	3.96* (0.16)	4.23* (0.81)
Accumulated income in high school (in log)	0.34* (0.02)	—
Predicted accumulated income in high school (in log)	—	0.37* (0.18)
Grades (in Z-score)	0.12* (0.04)	0.03 (0.06)
Parents income (in SEK 100,000)	0.01 (0.02)	−0.02 (0.02)
Years after graduation (t)	0.09* (0.01)	0.08* (0.01)
Birth quarter dummies	Yes	Yes
Cohort dummies	Yes*	Yes
Number of girls	3,220	1,447
Number of (repeated) observations	27,666	12,222

Note. Standard errors in parentheses.

*Indicates significant at 5% level.

As stated previously, the observational estimator is susceptible to selection bias. It comes natural to expect the estimator to overestimate the effect because some unobserved “ability” or “skill” is not accounted for. But, it is also possible that girls believing to have good academic prospects as well as good labor market prospects prefer to focus on their studies while in high school, in which case the observational estimator might underestimate the effect. To deal with this potential selection bias, we exploit the experimental characteristics of the data by using an Instrumental Variable approach (IV; see Angrist et al., 1996).

In this approach, the variable offered or not served as an instrument for the treatment (dose) variable.¹¹ The estimation method is essentially a two-stage least squares with the standard errors calculation being adjusted for the repeated measures of the post-schooling incomes. Technically, the estimation of the model is carried out as follows. Let the model matrix X be the row vectors $x_{i, t}^{T}$ stacked, $b = {(α, γ, δ, β_{1}, \dots, β_{k})}^{T}$ the vector of regression parameters, y the vector of the responses $l n (Y_{i, t})$ stacked, and Z a matrix identical to X except that the column of dose is replaced by the instrument. The IV estimator of b is computed as

\hat{b} = {(Z^{T} X)}^{- 1} Z^{T} y .

To account for the correlation of the error terms of Model (2), we use a generalized method of moments (GMM) estimator. The efficient GMM estimator gives the same parameter estimates in this case as the IV estimator but consistent standard errors (see Hayashi, 2000; pp. 204–215). The asymptotic covariance of $\hat{b}$ is given by

C o v (\hat{b}) = {({(Z^{T} X)}^{T} W^{- 1} (Z^{T} X))}^{- 1}

where $W = Z^{T} Σ^{- 1} Z$ and Σ the covariance matrix of the error terms of Model (2). In calculating the covariance matrix of Equation (4), Σ is replaced by an estimate of it using the residuals $e = y - X \hat{b}$ in accordance with Liang and Zeger (1986) and their Example 3.

Table 4 (right results column) shows that the post-schooling income increases with a point-estimate of 8% annually. Of most interest for this study is that the (log-) dose is significant with a point estimate of 0.37. The estimate suggests that a 1% increase in the accumulated income in high school causes 0.37% higher post-schooling incomes. One may wonder whether the effect wears off by time. To check that, we reestimated Model (2), including an interaction of dose and t: the statistically insignificant point estimate was −0.01 with a standard error of 0.02. Hence, the effect of work experience in high school seems to be strong and persistent later in the working life.

The estimation of Model (2) required us to make a number of arbitrary assumptions, which we afterward have checked with sensitivity analysis. First, we assumed a linear growth, δt_i in Model (2), in logged post-schooling income, but the estimated elasticity was unchanged when we included a quadratic term ( $δ_{1} t_{i} + δ_{2} t_{i}^{2}$ ) in the model. Second, the data set was incomplete as we did not have a 12-year follow-up after graduation for all the girls. A reestimation of the model using only a 5-year follow-up resulted in an estimated elasticity (standard error in parenthesis here and subsequently) of 0.40 (0.22). Third, the cohorts may have experienced different quality of treatment. Using only the cohorts of 1980–1983, we estimated the elasticity to be 0.27 (0.18). Fourth, immediately after graduation many of the girls had a low income or none. The explanation might be that they were studying at university. We reestimated the model using only post-schooling incomes beyond the university period (5 years after graduation) and found an elasticity of 0.37 (0.20). Moreover, we estimated the model for the post-schooling income of 2, 5, and 8 years after graduation and found elasticities of 0.31 (0.13), 0.48 (0.14), and 0.34 (0.16).

The program affected 283 girls that applied for a summer job for the first time as juniors. They were kept out of the analysis for methodological reasons as, by including them, there would be a risk for a downward bias of the estimate of the elasticity. Nonetheless, we included these 283 girls, reestimated Model (2), and found an elasticity of 0.31 (0.14). In this analysis, we defined the instrument variable offered as ever offered for all the girls having applied in either the sophomore or the junior year or both.

The sensitivity analysis also indicated a potential misspecification¹² of Model (2). It concerns the assumption of a homogenous effect with regard to the girls’ academic performance. Table 5 shows income while in high school, as well as post-schooling income for subsets of girls of varying academic achievements.

Table 5.

Treatment Effect and Academic Performance.

		Mean Difference in Logged Income for Offered vs. Nonoffered
	No. of Offered/Nonoffered	In High School	Years after Graduation
Subset: Girls with …			5	7
Grades in Z-score < 0	159/413	0.90 (0.16)	0.42 (0.24)	0.51 (0.27)
Grade in Z-score in (−1,1)	236/710	0.47 (0.11)	0.25 (0.10)	0.19 (0.20)
Grades in Z-score ≥ 0	203/672	0.39 (0.12)	0.14 (0.18)	0.02 (0.21)

Note. Standard errors (pooled) in parentheses.

The work experience in high school induced by the offer varied from an increase of about 90%, for the academically weaker girls, to about 40% for the girls, with above-average grades (Table 5). Following up on the girls 5 and 7 years after graduation, the girls who were offered summer jobs had higher incomes than those who were not offered jobs (ranging from 2% to 51%). The difference was most pronounced for the academically weaker girls. We reestimated Model (2) by including an interaction between dose and grade. An estimate, pertaining to a student with a grade of 0 on the Z-scale, of the elasticity was 0.32 (0.21). The estimate of the interaction effect was −0.1 (0.16). Hence, it cannot be ascertained that academically weak girls were benefitting the most from an early contact by the labor market. However, Knabe and Plum (2013) found less-skilled persons to benefit the most from low-paid work experience (cf. also Hensvik & Nordström-Skans, 2013).

As a consequence of the summer job program being voluntary and the potential heterogeneity in treatment effect, the interpretation of the elasticity for a wider group of Swedish high school students needs to be done with caution.

Conclusion

In this article, we addressed whether early contacts with the labor market, in the form of summer jobs for high school students, improved the transition from schooling to work. This issue is important since high youth unemployment rates prevail, and policies of this kind have been proposed and targeted to high school students in many countries. The empirical evidence on the issue is hard to find, particularly because the few existing studies are plagued by potential selection bias.

We studied the program effect of summer jobs offered by a Swedish municipality (Falun) under the umbrella of a national summer job program. In the program, the girls performed unskilled work related to elderly care, gardening, and cleaning, which are tasks routinely operated by the municipality. The duration of the job was set to 3 weeks and took place during the 9 weeks of school holidays in the summer.

What was the overall impact of the municipality’s summer job program on all high school students? We would say that the overall impact was minor. We have previously studied boys in the program and we found them less likely than girls to apply to the program, and if they did apply, they did not benefit from it (Alam, Carling, & Nääs, 2013). Furthermore, half of the girls did not apply to the program and the ones applying only had a 25% chance of being offered a summer job. Consequently, only 5% of high school students in Falun benefited directly from the program, with 19% higher post-schooling incomes. In our search for literature on general summer job programs for high school students, we found no previous study and therefore we cannot determine whether our results are typical for summer jobs programs.

Thus far, the municipality (or the government) has not considered targeting the program to subgroups of students. The fact that the summer jobs of the program were in the public sector further provides a possible explanation as to why (particularly academically weak) girls benefitted from the program. The public sector attracts mainly female workers and some of the girls in the program thereby acquired experience directly relevant for their future work tasks. In fact, we checked the proportion of girls with the municipality as employer the year after graduation from high school and found 37.0% for offered, 28.1% for nonoffered, and 28.4% for nonapplicants.

The program affected the girls’ accumulated work experience while in high school. We used the exogenous variation induced by the lottery to estimate the causal effect of accumulated income while in high school on post-schooling incomes, up to at the most 12 years after graduation. We found an elasticity of 0.37. Previous analysis by Hotz, Xu, Tienda, and Ahituv (2002) and Parent (2006) in the US and Canadian settings indicated a minor effect, whereas our finding is more in line with a relatively old nonexperimental study (Ruhm, 1997). Ruhm’s use of seniors, another definition of dose, and his focus on work while in high school rather than in the summer only render the comparison a bit difficult. However, the girls who were not offered any summer job had on average worked 26 weeks fulltime (SEK 65,000) by the year of graduation (see Table 3). Assuming that Ruhm’s 20 hr/week of work experience in the last year of high school amounts to adding 20 weeks of fulltime work experience to the existing 26, we then get an increase in dose of some 77% that would cause 31% higher post-schooling incomes (assuming a linear dose effect). Thus, Ruhm’s elasticity is similar to ours of 34%, when selection was controlled for only by means of background variables. These two estimates of the elasticity, however, are lower than ours of 37%, as found by means of the IV method. The effect also seems to be persistent, as we followed the girls to at most up to the age of 31, and there was no indication of the effect to wear off.

Interpreting our results as a general assessment of the importance of work experience while in high school should however be done with some care. In an attempt to check for heterogeneous effects, we found indications for the effect to be substantially higher for the girls with the weakest academic performance. The exogenous variation induced by the lottery primarily came from the girls with the weakest academic performance who are likely to directly opt for the labor market (of the public sector) upon graduation.

Footnotes

Acknowledgment

We thank seminar participants at Dalarna and the Linnaeus Universities, The Institute for Evaluation of Labor Market and Education Policy (IFAU), and the Swedish Ministry of Finance for constructive comments that improved this work. We are also grateful to Catia Cialani, Steven Haider, Iida Häkkinen Skans, Johan Vikström, Daniel Wikström, Olof Åslund, Siril Yella, Jacob Klerman, and the anonymous referees of ER for detailed comments on an earlier version of this article. This research was partially funded by IFAU.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was partially funded by the Institute for Evaluation of Labor Market and Education Policy (IFAU), Uppsala, Sweden.

Notes

References

Alam

Carling

Nääs

(2013). The effects of summer jobs on post schooling incomes. (Working Paper 2013:24). IFAU, Institute for Evaluation of Labour Market and Education Policy, Uppsala, Sweden.

Alon

Donahoe

Tienda

(2001). The effects of early work experience on young women’s labor force attachment. Social Forces, 79, 1005–1034.

Angrist

J. D.

Kruger

A. B.

(1991). Does compulsory school attendance affect schooling and earnings? The Quarterly Journal of Economics, 12, 979–1014.

Angrist

J. D.

Imbens

G. W.

Rubin

D. B.

(1996). Identification of causal effects using instrumental variables. Journal of the American Statistical Association, 91, 444–455.

Carling

Larsson

(2005). Does early intervention help the unemployed youth? Labour Economics, 12, 301–319.

Carr

R. V.

Wright

J. D.

Brody

C. J.

(1996). Effects of high school work experience a decade later: Evidence from the national longitudinal survey. Sociology of Education, 69, 66–81.

Geel

Backes-Gellner

(2012). Earning while learning: When and how student employment is beneficial. Labour, 26, 313–340.

Häkkinen

(2006). Working while enrolled in a university: Does it pay? Labour Economics, 13, 167–189.

Hayashi

(2000). Econometrics. Princeton, NJ: Princeton University Press.

10.

Hensvik

Nordström-Skans

(2013). Networks and youth labor market entry. (Working Paper 2013:23). IFAU, Institute for Evaluation of Labour Market and Education Policy, Uppsala, Sweden.

11.

Hotz

J. V.

C. L.

Tienda

Ahituv

(2002). Are there returns to the wages of young men from working while in school? Review of Economics and Statistics, 84, 221–263.

12.

Knabe

Plum

(2013). Low-wage jobs—springboard to high-paid ones? Labour, 27, 310–330.

13.

Lee

Orazem

P. F.

(2010). High school employment, school performance, and college entry. Economics of Education Review, 29, 29–39.

14.

Leos-Urbel

(2012). What is a summer job worth? The impact of summer youth employment on short-term academic outcomes: Evidence from a large-scale lottery. (Working Paper 5:2012, IESP Working Paper Series) Institute for Education and Social Policy, New York University, New York.

15.

Liang

K.-Y.

Zeger

S. L.

(1986). Longitudinal data analysis using generalized linear models. Biometrika, 73, 13–22.

16.

Littell

R. C.

Milliken

G. A.

Stroup

Wolfinger

R. D.

(1996). SAS system for mixed models. Cary, NC: SAS Institute Inc.

17.

Monahan

K. C.

Lee

J. M.

Steinberg

(2011). Revisiting the impact of part-time work on adolescent adjustment: Distinguishing between selection and socialization using propensity score matching. Child Development, 82, 96–112.

18.

Morisi

T. L.

(2010) May. The early 2000s: A period of declining teen summer employment rates. US Bureau of Labor Statistics, Monthly Labor Review, 133, 23–35.

19.

OECD. (2014). Education at a glance 2014: OECD indicators. OECD Publishing, Paris. Retrieved October 15, 2014, from http://www.oecd-ilibrary.org/education/education-at-a-glance_19991487

20.

Parent

(2006). Work while in high school in Canada: Its labour market and educational attainment effects. Canadian Journal of Economics, 39, 1125–1150.

21.

Patton

Smith

(2010). Part-time work of high school students: Impact on employability, employment outcomes and career development. Australian Journal of Career Development, 19, 54–62.

22.

Rosenbaum

J. E.

DeLuca

Miller

S. R.

Roy

(1999). Pathways into work: Short-and long-term effects of personal and institutional ties. Sociology of Education, 72, 179–196.

23.

Ruhm

C. J.

(1997). Is high school employment consumption or investment? Journal of Labour Economics, 15, 735–776.

24.

Staff

Schulenberg

J. E.

Bachman

J. G.

(2010). Adolescent work intensity, school performance, and academic engagement. Sociology of Education, 83, 183–200.

25.

Stock

J. H.

Watson

M. W.

(2003). Introduction to Econometrics. Boston, MA: Addison Wesley.

26.

UKÄ. (2013). Higher education in Sweden: 2013 status report. Stockholm, Sweden: Universitetskanslersämbetet (UKÄ, in English: Swedish Higher Education Authority). Retrieved October 15, 2014, from www.uk-ambetet.se

27.

Wang

Carling

Nääs

(2006). High school students’ summer jobs and their ensuing labour market achievements. (Working Paper 2006:14). IFAU, Institute for Labour Market Policy Evaluation, Uppsala, Sweden.

28.

Weller

N. F.

Cooper

S. P.

Basen-Engquist

Kelder

S. K.

Tortolero

S. R.

(2003). School-year employment among high school students: Effects on academic, social and physical functioning. Adolescence, 38, 441–459.