- Email: [email protected]
Career trajectory in high school dropouts
The Social Science Journal 50 (2013) 306–312
Contents lists available at SciVerse ScienceDirect
The Social Science Journal journal homepage: www.elsevier.com/locate/soscij
Career trajectory in high school dropouts Kyung-Nyun Kim ∗ Korea Research Institute for Vocational Education & Training, Kangnam-Gu, Chongdam-dong 15-1, Seoul 135-949, South Korea
a r t i c l e
i n f o
Article history: Received 25 July 2012 Received in revised form 11 March 2013 Accepted 12 March 2013 Available online 8 April 2013
Keywords: Career trajectory High school dropouts Individual trait Latent class growth analysis
a b s t r a c t This study considers the career trajectories of high school dropouts, which has been given little attention to in the literature. Considering worker heterogeneity for individuals who do not complete high school, we estimate possible career trajectories and investigate the traits related with the decision to drop out. Using latent class growth analysis, three kinds of career trajectories are identified: dead-end, stepping-stone, and advancing careers. Although the majority of dropouts are in the dead-end careers, about 30% are in the process of escaping low-status jobs through acquiring work experience. Individual traits, such as gender, race, and cognitive ability, as well as home computer access are significantly related to the different types of career trajectories. © 2013 Western Social Science Association. Published by Elsevier Inc. All rights reserved.
1. Introduction Every year, 1.2 million students in the United States leave high school without a diploma (Alliance for Excellent Education, 2011). Dropping out of school not only limits one’s life-time opportunities, it also creates a social cost. Legislation and social programs have been implemented to keep kids in school and increase their employment opportunities (Campolieti, Fang, & Gunderson, 2010; Kuenzi, 2007). However, later life opportunities for high school dropouts remain meager, and many become welfare recipients (Alliance for Excellent Education, 2011; Danziger et al., 1999; Schwartz, 1995). Existing studies assume a homogenous population of high school dropouts; however, this population is likely heterogeneous (Brown, 1982; Cluck, Beaulieu, & Barfield, 1998; Waldinger & Lichter, 2003). However, one fact that cannot be overlooked is that not every high school dropout ends up in a dead-end job. Kusmin and Gibbs (2000) demonstrate that one fifth of initial entry jobs held by less-educated workers can lead to subsequent better
∗ Tel.: +82 10 8735 1184; fax: +82 2 3485 5140. E-mail address: [email protected]
paying jobs, implying that some dropouts acquire jobs that require at least a high school diploma or beyond; hence, some earn enough money to support themselves without public assistance. Kusmin and Gibbs (2000) identify initial jobs that move less-educated workers up occupation ladders into better paying jobs. However, few studies attempt to identify career trajectories of US high school dropouts, and it is important to identify the occupational trajectories of these less-educated workers if there is to be successful intervention strategies. Other dropouts located at the bottom of the occupational hierarchy require an occupational intervention that complements their low education levels to help them move into middle-level occupations. Moreover, finding the traits related to lower occupational career patterns better orients policies that create different opportunities for the status of dropouts. The aim of this study is, therefore, to identify how high school dropout career trajectories are distributed and what dropout characteristics contribute to the heterogeneous career trajectories. 2. Career mobility among high school dropouts Several studies demonstrate the later-life occupations held by dropouts in dead-end (Brown, 1982) and secondary jobs (Waldinger & Lichter, 2003). In dead-end jobs, work
0362-3319/$ – see front matter © 2013 Western Social Science Association. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.soscij.2013.03.005
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
experience increases wages by only a little (Connolly & Gottschalk, 2002); hence, dropouts in dead-end jobs are resigned to sub-standard career development. Studies on dead-end jobs emphasize that dropouts sort into lowstatus jobs according to little human capital, and their careers are likely stuck in lower occupational statuses if they do not enhance their skills and knowledge (Farkas, England, Vicknair, & Kilbourne, 1997; Mouw & Kalleberg, 2010). However, some jobs, even for dropouts, can be stepping stones to higher status occupations (Kusmin & Gibbs, 2000). One approach is work-first, which emphasizes that work experience accumulation may lead dropouts to advance up occupational ladders regardless of their entry level (Holcomb, Pavetti, Ratcliffe, & Riedinger, 1998). US high school dropouts have more frequent turnover and have less attachment to the labor market. For dropouts, high worker turnover can be improved with early employment intervention (Daguerre, 2007). 3. Factors leading to occupational mobility Studies show that occupational mobility differs by demographics, including gender, race, and parental background (Grusky, 2008). Moreover, women are disadvantaged more than men from lower occupational status due to occupational segregation (Sewell, Hauser, & Wolf, 1980), lower occupational mobility (Topel & Ward, 1992), and labor market discrimination (Maume, 1999). However, opportunities are changing with manufacturing off-shoring, which allows workers to change jobs from the manufacturing to the service industry (Schrammel, 1998). With this off-shoring trend, male dominated occupations have become scarce (Schrammel, 1998) and pushed male workers into lower status occupations compared with women, because women are more skilled with service sector skills. Minorities continue to be discriminated against (Grusky, 2008); there are ability differences (Herrnstein & Murray, 1994) and residential segregation persists (Wilson, 1996). Different distributions also exist for labor market information, and social capital differences between families may have an impact on dropout career development. Parental background through social capital on children’s occupational outcomes creates heterogeneity across dropouts and less-educated and single parents have fewer resources, which also creates unobserved variation across dropouts (Evans, Kelly, Sikora, & Treiman, 2005; Lin, Ensel, & Vaughn, 1981). Studies also demonstrate that occupational mobility differs by individual traits, such as human capital and self-esteem. First, efficiency differences and investment effectiveness by cognitive ability create different opportunities that are related to the occupational status of dropouts (Farkas et al., 1997). Workplace skill-biased technological change (SBTC) leads to increased demands for skilled workers, which accentuate the relationship between cognitive ability and wages (McCall, 2000). SBTC also favors worker computer use skills, which results in higher wage returns (Autor, Katz, & Krueger, 1998; Hotchkiss & Shiferaw, 2011). Moreover, the workplace digital divide is related to computer use at home (Pliskin, Levy, Heart, O’Flaherty, & O’Dea,
307
2006). Studies also demonstrate self-esteem as a labor market advantage, and high self-esteem workers are more likely to hold high status (Heckman & LaFontaine, 2006; Rosenberg & Owens, 2001). Workers with low self-esteem may take on a low status jobs (Gottfredson, 1981; McNatt & Judge, 2004). These unobserved characteristics have not been considered for US dropouts and their later life job opportunities. 4. Methods 4.1. Data Data from the National Longitudinal Study Youth 97 (NLSY 97) 1997 to 2008 are used to consider career trajectories of US high school dropouts. The NLSY 97 is a longitudinal survey of 8984 respondents aged 12–16 as of December 31, 1996 (U.S. Department of Labor, 2009). Respondents in NLSY 97 are interviewed each year for information about school-to-work transitions, including schooling and participation in the labor market. NLSY 97 consists of two samples: a cross-sectional sample that represents the U.S. population born between 1980 through 1984, and a supplemental sample of black and Hispanic populations born in the same years. Sample weights are applied to adjust for differences in the probability of selection due to minority oversampling. To investigate traits related to dropout occupational standing, we limit the sample to those who dropped out of high school by 2003. This is the expected graduation year if 12 year old respondents as of December 31, 1996 finished high school in 4 years. This decreases the number of observations from 8984 to 1813. Moreover, 438 respondents are excluded who enrolled in college after acquiring a GED by 2008, which is the year of the latest release of NLSY 97. This decreases the sample size to 1555. 4.2. Dependent variable 4.2.1. Occupational career Occupational standing over time is used as the dependent variable, which has information on “a degree of prestige or social standing” (Miech, Eaton, & Liang, 2003, p. 441). Ever since Duncan developed socioeconomic index (SEI), many scholars have updated his SEI. This study uses updated Duncan’s SEI measure by Nakao and Treas (1994). Occupations are those respondents held at the beginning of each year. 4.3. Independent variables 4.3.1. Years since high school dropout Years since high school dropout is used for the time scale variable, which is the baseline of the growth model in occupational standing. Years since high school dropout are a proxy for work experience. Due to missing data in work history, it is difficult to use the actual cumulative work experience in the growth model. This study uses Mincer’s approach (1974) for work experience, which is age minus years of schooling. Mincer’s approach measures the
308
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
potential work experience, which assumes a continuous work history. 4.3.2. Demographic variables This study considers five demographic variables: family structure, parental education, gender, language at home, and race. Family structure is classified into three groups: single- and two-parent households and other types of households, such as those headed by grandparents or children. Two-parent households are the reference group. Parental education is coded as a series of dummy variables: less than high school, 2nd year college, and 4th year college or above. High school is the reference for parental education. Home language other than English is dummy coded with the reference of English speaking. Race is dummy coded for black and Hispanic. White is the reference for race. 4.3.3. Computer at home Computer at home measures whether respondents had access to a computer at home during their high school years and is dummy-coded. No computer at home is the reference. 4.3.4. Cognitive ability Two cognitive ability variables are used: ArmedServices Vocational Aptitude Battery (ASVAB) and grade completed. ASVAB is the percentile scores measured at ages 13–15 for four tests measuring verbal and mathematical ability: arithmetic reasoning, mathematical knowledge, word knowledge, and paragraph comprehension. Ageadjusted ASVAB scores are used to adjust for a cohort difference at the time of measurement. Grade completed is the total years in school before dropping out. 4.3.5. Self-esteem Self-esteem in NLSY 97 is measured with the Rosenberg scales (Rosenberg, 1965). A mean index of the self-esteem scale, which is composed of the following four items: (1) I hardly ever expect things to go my way, (2) I rarely count on good things happening to me, (3) I am always optimistic about my figure, (4) in uncertain times, I usually expect the best. The items are scaled at four levels: 1 – strongly disagree, 2 – disagree, 3 – agree, and 4 – strongly agree. The first two negative items are coded for higher scores to reflect higher self-esteem (Cronbach’s ˛ = 0.68). 4.4. Data analyzes 4.4.1. Latent class growth analysis (LCGA) The latent class growth analysis (LCGA) is employed to classify a number of subgroups with heterogeneous trajectories as well as to identify the shape of subgroup trajectories, and the purpose of LCGA is to classify a population into an unknown set of latent classes (Duncan, Susan, Strycker, & Okut, 2002; Vaughn & Witko, 2012). Moreover, assuming that a single growth model specifies a population, standard latent growth analysis produces a single mean trajectory of all individuals in a sample (Jung & Wickrama, 2008). However, a single model may offset different trends
between heterogeneous groups (Nagin, 2005). LCGA models several trajectories of homogeneous workers within groups by classifying dropouts who show similar patterns of occupational trajectories and those who can be grouped together (Nagin, 1999). Applying LCGA allows for heterogeneous career trajectories of dropouts. LCGA analyzes continuous outcomes measured repeatedly over time. Trajectory classes in LCGA are defined as a set of individuals who follow a unique pattern (Luyckx, Schwartz, Goossens, Soenens, & Beyers, 2008). Thus, the growth parameters in LCGA, such as the intercept and slopes, can be different between groups. LCGA is a groupbased statistical technique and a model fit index classifies individuals into groups. LCGA is conducted using MPlus. 4.4.2. Model fit index To determine the number of latent trajectory classes, three different types of model fit indexes are used, such as Bayesian information criteria (BIC), Lo–Mendell–Rubin (LMR), and entropy (Jung & Wickrama, 2008). The model with the smallest BIC [BIC = (−2 log likelihood + p log(n)), where p is the number of parameters and n is the sample size] is preferred. LMR tests, which compare k class model and k − 1 class model, provide a model fitting better the data. Significant LMR value indicates that a model with k class is preferred to a model with k − 1 class. The closer the entropy value is to one, the better the model performs.1 4.4.3. Multinomial logistic regression After selecting the class number, a multinomial logistic regression model is estimated to identify variables predicting class membership. To identify how dropout traits are associated with occupational advancement, a multinomial logistic model is used on the dropout data. 4.4.4. Missing data Missing data is imputed for individual characteristics, such as cognitive ability and parental education using the MI command in Stata. 5. Results 5.1. Descriptive statistics Descriptive statistics are in the second column of Table 1. There are more male respondents (58%) than female respondents (42%). Fifty-eight percent of the respondents are white, while black, Hispanics, and other minorities represent 21%, 8%, and 13% of the sample, respectively. About 48%, 34%, and 27% are from two-parent, single-parents, and no-parents family, respectively. Furthermore, approximately 33% of parents have less than high school, 48% have only high school, 12% have 2nd year college, and 7% of the respondents have parents with 4 year
1 Since LCGA models the patterns of change in a dependent variable across multiple time intervals, it differs from an analytic technique for finding group differences common with analysis of variance or analyzing changes between two discrete time intervals, such as t-tests (Curran and Muthen, 1999).
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
309
Table 1 Descriptive statistics and distribution of three-trajectory class by individual traits. Variable
Overall
Men Black Hispanic Other-race Single-parent Other-parent Less than HS 2nd year college 4th year college Other language at home Computer at home Grades completed (M) ASVAB (M) Self-esteem (M)
.58 (.49) .21 (.41) .08 (.26) .13 (.33) .34 (.47) .27 (.44) .33 (.48) .12 (.34) .07 (.26) .17 (.39) .62 (.49) 9.91 (2.72) 31.32 (23.31) 2.73 (.34)
Three-trajectory class Dead-end
Stepping-stone
Advancing
F
.59 (.49)a .25 (.43)a .07 (.26) .12 (.33) .35 (.48)a .26(.44) .34 (.48) .11 (.33) .06 (.25) .17 (.39) .57 (.50)a 9.72 (1.22)a 29.62 (19.54)a 2.75 (.34)
.58 (.49)a .13 (.33)b .09 (.28) .14 (.35) .25 (.46)b .31 (.46) .31 (.47) .14 (.36) .08 (.29) .15 (.37) .70 (.47)b 10.21 (2.09)b 34.02 (19.92)b 2.72 (.34)
.38 (.49)b .17 (.38)a .08 (.28) .14 (.35) .45 (.49)a .28 (.49) .31 (.46) .11 (.37) .07 (.33) .18 (.37) .74 (.44)b 10.11 (1.16)b 35.61 (22.26)b 2.74 (.30)
** **
**
** + *
Source: The National Longitudinal Study Youth 97. Notes: Descriptive statistics with different subscripts differ. M stands for mean. + p < .07. * p < .05. ** p < .01.
college degrees. Seventeen percent of the sample speaks a language other than English at home, and 62% had a computer at home during high school. The average grade completed is 9.91. The average percentile score for ASVAB is 31.32, and the average score for self-esteem is 2.73.
5.2. Identifying the number of classes Model estimation begins with a normal latent growth model, and respondents have different time spans according to their cohort membership at time of dropout. A cohort-sequential design is used to estimate the growth model. To identify the time dimension variable, a likelihood test is conducted between linear and quadratic models. The difference in the likelihood statistics between the two models is significant [2 (56, 4) = .000, p < .001], indicating that a quadratic model better fits the data than a liner model. However, in the latter process of identifying a latent class, a quadratic factor is statistically insignificant. Thus, a linear model is specified for further analysis. An unconditional latent cluster model is then specified by constraining growth factor variance and covariance. This assumes the intercept and growth rates are zero. These constraints make individual growth trajectories homogenous within the same class (Jung & Wickrama, 2008). Model fit statistics determines the number of latent classes. Different models are tested by increasing the numbers of clusters. Table 2 presents model fit statistics for cluster models 1 through 4. To determine the number of classes, model fit indexes, such as BIC, entropy and LMR, are used. In applying the model fit index, Kusmin and Gibbs’ finding (2000) is used, which suggests that most high school dropouts would be stuck in dead-end jobs; however, a few overcome the low initial job placement associated with leaving high school without a degree. As a result, most dropouts fall into lower-stable trajectories, whereas a few reach increasing trajectories. The BIC index, indicating that higher values are a better fit, supports a four-class solution. However,
entropy supports the two-class solution indicating that higher values accurately classify individuals into latent classes based on the posterior probability classification. LMR demonstrates indicates that current class inclusion leads to a better fit, and supports the three-class solution. Between two- and three-class solutions, BIC and LMR favor the three-class solution over the two-class solution. The three-class model places individuals into each trajectory with 88%, 80%, and 90% accuracy. Fig. 1 displays fitted growth models for the three-class solution. Class 1 is individuals who began at lower occupation levels and remained in low status jobs. Class 1, is about 69% of the sample and depicts dead-end careers. Class 2 shows a separate career trajectory compared with class 1 and begins at a low occupation level, but increases occupational status over time. Class 2 represents stepping-stone
Fig. 1. Fitted growth lines of occupational status for three-trajectory classes. Source: The National Longitudinal Study Youth 97.
310
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
Table 2 Identifying the number of trajectory class. No. of classes
Model fit index
Posterior probability for 3-class model
BIC
Entropy
LMR
1
2
3
1 2 3 4
57,066 55,345 55,056 54,948
– .90 .72 .67
– .00 .05 .70
.883 .037 .004
.115 .803 .097
.001 .160 .902
Proportions of latent classes
.694 .231 .075
Source: The National Longitudinal Study Youth 97. Notes: Three-class model is selected. Analyzes are conducted using Mplus.
careers and is around 23% of the sample. Class 3 begins with a moderate occupation level and occupational careers increases over time. Class 3 is around 8% of the sample, represents advancing careers, and indicates that low educational qualifications do not influence occupational status. Thus, a majority of high school dropouts end up with deadend careers (69%), while about 31% of them experience career advancement over time. Each class’ estimated growth factor is presented in Table 3. Stepping-stone and advancing careers have increasing occupational growth trajectories, whereas dead-end careers and stepping-stone careers have similar initial occupational status. The intercept and slopes for the three classes demonstrate that for stepping-stone careers, both the intercept and slope are significant. Initial occupational status is 35.78, and the growth rate is 1.11. Stepping-stone careers of dropouts predict that workers hold onto jobs located at the median of the occupation scale for 10 years after high school dropout. For dead-end careers, only the intercept is significant and their initial occupational status is 33.47 with a growth rate of −0.08. The predicted occupational status for 10 year after dropout is 33.47; hence, dropouts in this career remain at the bottom of the occupational distribution. For advancing careers, initial occupational status is 50.52, and the growth rate is 1.15. The predicted value of occupational status 10 years after high school dropout is 62, and the predicted occupational status among different career types 10 years after dropout is considerable. 5.3. Identifying traits related to class trajectory membership Table 1 columns 3–6 present cross-tabulations of trajectory classes according to dropout traits. A one-way analysis of variance with adjustment for alpha levels demonstrates Table 3 Estimates for growth factors by trajectory class. Trajectory class
1. Dead-end careers 2. Stepping-stone careers 3. Advancing careers Total sample
significant class composition differences by gender, black, single parent, computer at home, grades completed in high school, and ASVAB. A multinomial logistic regression is estimated to examine traits related to class trajectory membership, and the reference category is dead-end career trajectory. Dead-end careers are used as the denominator, therefore, contrasts with other trajectory patterns that allow for testing for the traits related to dropouts’ occupational advancement.2 Traits that effect trajectory membership are gender, black, computer at home, and cognitive ability. Family structure, parental education, other language at home, and self-esteem are not significant. Table 4 includes multinomial logistic regression coefficients as logged odds, odds ratios, and confidence intervals. For male dropouts, the odds of being in dead-end careers are 0.40 times the odds for being in an advancing career, indicating that male dropouts are less likely of being in advancing careers. Men are more likely to be in a deadend career than an advancing career. The odds of blacks being in a stepping-stone career are 0.41 times that of a dead-end career, which means that among the cohort of dropouts that blacks are less likely to advance occupationally. There is, however, no difference in the composition of blacks in dead-end careers and advancing careers. If the initial occupational status of black workers is low, they stay in lower job tracks. However, if equipped with a computer at home, dropouts are more likely to be in stepping-stone careers or in advancing careers. The odds of being in stepping-stone and advancing careers for dropouts with a computer in the home during high school are 1.61 and 2.26 the odds for being in a dead-end career, which means they are 61% and 126% more likely of being in stepping-stone and career advancing careers later in life if they have a computer in the home during high school. Finally, the odds of being in a stepping-stone career with a one unit increase in cognitive ability are 1.01 times that of being in a dead-end career. Therefore, workers with greater cognitive abilities are more likely to be in stepping-stone than in dead-end careers.
Growth factor estimate Mean intercept
Mean linear slope
6. Discussion
33.47*** 35.78*** 50.52*** 35.41***
−.08 (.10) 1.11*** (.31) 1.15** (.54) .34*** (.07)
This study considers career trajectories of high school dropouts. Given dropout heterogeneity, the types of career
(.55) (1.13) (4.11) (.41)
Source: The National Longitudinal Study Youth 97. Notes: Analyzes are conducted using Mplus. ** p < .01. *** p < .001.
2 Multinomial logistic regression can be denoted by ln(Ps/Pd) and ln(Pa/Pd), where Ps, Pa, and Pd represent the probabilities of being in stepping-stone, advancing, and dead-end careers.
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
311
Table 4 Results of multinomial logistic model for trajectory class membership. Variable
ˇ (stepping-stonea )
eˇ (95% CI)
ˇ (advancinga )
eˇ (95% CI)
Men White (reference) Black Hispanic Other race Dual-parent (reference) Single-parent Other-parent High school (reference) Less than HS 2nd year college 4th year college Other language at home Computer at home Grades completed ASVAB Self-esteem Constant
−.11 (.13)
.89 (.69; 1.15)
−.74** (.21)
.48 (.32; .72)
−.51** (.18) .23 (.22) .13 (.21)
.59 (.42; .85) 1.26 (.03; .82) 1.14 (.04; .75)
−.19 (.30) .15 (.40) .04 (.34)
.83 (.46; 1.49) 1.16 (.53; 2.54) 1.04 (.53; 2.03)
−.25 (.16) .09 (.17)
.78 (.57; 1.07) 1.09 (.78; 1.53)
.39 (.26) .43 (.29)
1.48 (.89; 2.46) 1.54 (.87; 2.71)
.03 (.16) .19 (.22) .18 (.28) −.18 (.19) .51** (.14) .06 (.05) .01+ (.00) −.02 (.18) −1.95** (.75)
1.03 (.75; 1.41) 1.21 (.79; 1.86) 1.20 (.69; 2.07) .84 (.58; 1.21) 1.61 (1.15; 2.25) 1.06 (.96; 1.17) 1.01 (.99; 1.02) .98 (.69; 1.39) .14 (.03; .62)
−.18 (.27) −.30 (.37) .09 (.41) .43 (.28) .82** (.24) .06 (.06) .01 (.00) .24 (.43) −4.14** (1.24)
.84 (.49; 1.42) .74 (.36; 1.53) 1.09 (.49; 2.44) 1.54 (.89; 2.66) 2.26 (1.42; 3.63) 1.06 (.94; 1.19) 1.01 (1.01; 1.02) 1.27 (.55; 2.95) .02 (.00; .18)
Model fit
F(30, 6015) = 3.16, p < .001
Source: The National Longitudinal Study Youth 97. Notes: Analyzes are conducted using Mplus. a Reference is dead-end career. + p < .07. ** p < .01.
trajectories that exist for dropouts and their traits are related to trajectory membership. Using latent class growth analysis, three kinds of career trajectories are identified: dead-end career, stepping-stone career, and advancing career. While the majority of dropouts are in the dead-end careers, about 31% are in the process of escaping low-status jobs through accumulating work experience. Individual traits, such as gender, race, and cognitive ability, and having a computer in the home at early ages, are significantly related with different career trajectories. A majority of dropouts (69%) are trapped in low-status jobs and thus have little mobility. For these workers, acquiring work experience provides no benefit. About 23% of dropouts start out in low-status jobs; however, they escape mundane entry-level jobs and move into better jobs, and initial experience from low-status jobs becomes a stepping stone to a better position. About 8% of dropouts, despite their academic disadvantage, start at middle-status jobs and develop their careers. The traits of dropouts are related to their career trajectories. Despite institutional improvements for different ethnic groups, black dropouts are more likely end up in dead-end careers relative to their white counterparts. However, other minorities are not disadvantaged in their career trajectories. Second, a computer in the home during school is related to the likelihood of obtaining a stepping-stone or an advancing career. The benefit of having a computer at home prepares workers for the modern labor market, and the lack of computer skills may drive dropouts to more rudimentary jobs that do not offer career advancement. Third, for workers at low-level jobs, cognitive ability continues to shape their life opportunities. This may indicate that intelligent workers acquire required skills more easily and allow them to move out of dead-end jobs with greater work experience. Fourth, gender-based
occupational differentials for dropouts are reversed. Female workers are more likely to follow advancing careers than their male counterparts, which is consistent with prior studies that gender wage gaps narrow (Holzer & Hlavac, 2011). The advantage of being female among less-educated workers may be attributable to the change in an industrial mix, and a growing service sector but declining manufacturing sector. References Alliance for Excellent Education. (2011). The high cost of high school dropouts: What the nation pays for inadequate high schools. Retrieved from:. http://www.all4ed.org/files/HighCost.pdf Autor, D., Katz, L., & Krueger, A. (1998). Computing inequality: Have computers changed the labor market? The Quarterly Journal of Economics, 113(4), 1169–1213. Brown, C. (1982). Dead end jobs and youth unemployment. In R. Freeman, & D. Wise (Eds.), The youth labor market problem (pp. 427–452). Chicago: University of Chicago Press. Campolieti, M., Fang, T., & Gunderson, M. (2010). Labor market outcomes and skill acquisition of high-school dropouts. Journal of Labor Research, 31(1), 39–52. Cluck, R. E., Beaulieu, L. J., & Barfield, M. A. (1998). To the educated, the spoils: The relation of education to labor market experiences of young adults (ERIC Document Reproduction Service No. ED432418). Connolly, H., & Gottschalk, P. (2002). Job search with heterogeneous wage growth transitions to “better” and “worse” jobs. Retrieved from:. http://www.economics.neu.edu/papers/documents/03-003.pdf Curran, P. J., & Muthen, B. O. (1999). The application of latent curve analysis to testing developmental theories in intervention research. American Journal of Community Psychology, 27, 567–595. Daguerre, A. (2007). Active labor market policies and welfare reform. Basingstoke, Hants: Palgrave. Danziger, S. K., Corcoran, M., Danziger, S., Heflin, C., Kalil, A., & Levine, J. (1999). Barriers to the employment of welfare recipients. Retrieved from:. http://www.psc.isr.umich.edu/pubs/pdf/rr02-508.pdf Duncan, T. E., Susan, S. C., Strycker, L. A., & Okut, H. (2002). Growth mixture modeling of adolescent alcohol use data: Chapter addendum to An Introduction to Latent Variable Growth Curve Modeling: Concepts, Issues, and Applications. Eugene, OR: Oregon Research Institute.
312
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
Evans, M., Kelly, J., Sikora, J., & Treiman, D. (2005). To the scholar go the spoils? The influence of parents’ scholarly culture on offspring’s occupational attainment in 31 countries. Paper presented at the annual meeting of International Sociological Association, Los Angeles, CA. Farkas, G., England, P., Vicknair, K., & Kilbourne, B. (1997). Cognitive skill, skill demands of jobs, and earnings among young European-American, African-American, and Mexican-American Workers. Social Forces, 75, 913–940. Gottfredson, L. S. (1981). Circumscription and compromise: A developmental theory of occupational aspirations. Journal of Counseling Psychology, 28, 545–579. Grusky, D. B. (2008). Social stratification: Class, race, and gender in sociological perspective (3rd ed.). Boulder, CO: Westview Press. Heckman, J. J., & LaFontaine, P. (2006). Bias corrected estimates of GED returns. Journal of Labor Economics, 24, 661–700. Herrnstein, R. J., & Murray, C. (1994). The bell curve: Intelligence and class structure in American life. New York, NY: The Free Press. Holcomb, P. A., Pavetti, L., Ratcliffe, C., & Riedinger, S. (1998). Building an employment focused welfare system: Work first and other work-oriented strategies in five states. Washington, DC: The Urban Institute. Holzer, H. J., & Hlavac, M. (2011). An uneven road and then a cliff: U.S. labor markets, 2000–2010. Retrieved from:. http://www.s4.brown. edu/us2010/projects/authors lm.htm Hotchkiss, J. L., & Shiferaw, M. (2011). Decomposing the education wage gap: Everything but the kitchen sinks. Federal Reserve Bank of St. Louis Review, July, 243–272. Jung, T., & Wickrama, K. (2008). An introduction to latent class growth analysis and growth mixture modeling. Social and Personality Psychology Compass, 2, 302–317. Kuenzi, J. (2007). High school graduation, completion, and dropouts federal policy, programs, and issues (RL-33963). Retrieved from:. http://www.lieberman.senate.gov/assets/pdf/crs/highschool.pdf Kusmin, L., & Gibbs, R. (2000). Less-educated workers face limited opportunities to move up to good jobs. Rural America, 15(3), 33. Lin, N., Ensel, W. M., & Vaughn, J. C. (1981). Social resources and strength of ties: Structural factors in occupational status attainment. American Sociological Review, 46, 393–405. Luyckx, K., Schwartz, S. J., Goossens, L., Soenens, B., & Beyers, W. (2008). Developmental typologies of identity formation and adjustment in female emerging adults: A latent class growth analysis approach. Journal of Research on Adolescence, 18, 595–619. Maume, D. J., Jr. (1999). Glass ceilings and glass escalators: Occupational segregation a race and sex differences in managerial promotions. Work and Occupations, 26(4), 483–509. McCall, L. (2000). Explaining levels of within-group wage inequality in U.S. labor market. Demography, 37, 415–430. McNatt, D. B., & Judge, T. A. (2004). Boundary conditions of the Galatea effect: A field experiment and constructive replication. Academy of Management Journal, 47, 550–565.
Miech, R. A., Eaton, W., & Liang, K. Y. (2003). Occupational stratification over the life course: A comparison of occupational trajectories across race and gender during the 1980s and 1990s. Work and Occupations, 30, 440–473. Mincer, J. (1974). Schooling, earnings, and experience. New York: Columbia University Press. Mouw, T., & Kalleberg, A. L. (2010). Stepping stone versus dead end jobs: Occupational pathways out of working poverty in the United States, 1996–2007. Retrieved from:. http://www.unc. edu/∼tedmouw/papers/stepping%20stone%20versus%20dead%20end %20jobs%2008102010.pdf Nagin, D. (1999). Analyzing developmental trajectories: A semiparametric, group-based approach. Psychological Methods, 4(2), 139–157. Nagin, D. S. (2005). Group-based modeling of development. Cambridge, MA: Harvard University Press. Nakao, K., & Treas, J. (1994). Updating occupational prestige and socioeconomic scores: How the new measures measure up. In P. Marsden (Ed.), Sociological methodology (pp. 1–72). Washington, DC: American Sociological Association. Pliskin, N., Levy, M., Heart, T., O’Flaherty, B., & O’Dea, P. (2006). The corporate digital divide between smaller and larger firms. In E. Trauth, D. Howcroft, T. Butler, B. itzgerald, & J. DeGross (Eds.), Social inclusion: Societal and organizational implications for information systems (pp. 413–417). Boston, MA: Springer. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton: Princeton University Press. Rosenberg, M., & Owens, T. (2001). Low self-esteem people: A collective portrait. In T. J. Owens, S. Stryker, & N. Goodman (Eds.), Extending self-esteem theory and research (pp. 400–436). New York: Cambridge University Press. Schrammel, K. (1998). Two generations of young adults: Comparing the labor market success of young adults from two generations. Monthly Labor Review, February, 3–9. Schwartz, W. (1995). School dropouts: New information about an old problem. Retrieved from:. http://www.ericdigests.org/ 1996-2/dropouts.html Sewell, W. H., Hauser, R., & Wolf, W. (1980). Sex, schooling, and occupational status. American Journal of Sociology, 863, 551–583. Topel, R. H., & Ward, M. P. (1992). Job mobility and the careers of young men. The Quarterly Journal of Economics, May, 439–479. U.S. Department of Labor. (2009). NLSY97 user’s guide. Retrieved from:. http://www.nlsinfo.org/nlsy97/nlsdocs/nlsy97/maintoc.html Vaughn, M., & Witko, C. (2012). Does the amount of school choice matter for student engagement? The Social Science Journal, http://dx.doi.org/10.1016/j.bbr.2011.03.031 [Advance online publication] Waldinger, R., & Lichter, M. (2003). How the other half works: Immigration and the social organization of labor. Los Angeles, CA: University of California Press. Wilson, W. (1996). When work disappears: The world of the new urban poor. New York: Alfred A. Knopf.
Contents lists available at SciVerse ScienceDirect
The Social Science Journal journal homepage: www.elsevier.com/locate/soscij
Career trajectory in high school dropouts Kyung-Nyun Kim ∗ Korea Research Institute for Vocational Education & Training, Kangnam-Gu, Chongdam-dong 15-1, Seoul 135-949, South Korea
a r t i c l e
i n f o
Article history: Received 25 July 2012 Received in revised form 11 March 2013 Accepted 12 March 2013 Available online 8 April 2013
Keywords: Career trajectory High school dropouts Individual trait Latent class growth analysis
a b s t r a c t This study considers the career trajectories of high school dropouts, which has been given little attention to in the literature. Considering worker heterogeneity for individuals who do not complete high school, we estimate possible career trajectories and investigate the traits related with the decision to drop out. Using latent class growth analysis, three kinds of career trajectories are identified: dead-end, stepping-stone, and advancing careers. Although the majority of dropouts are in the dead-end careers, about 30% are in the process of escaping low-status jobs through acquiring work experience. Individual traits, such as gender, race, and cognitive ability, as well as home computer access are significantly related to the different types of career trajectories. © 2013 Western Social Science Association. Published by Elsevier Inc. All rights reserved.
1. Introduction Every year, 1.2 million students in the United States leave high school without a diploma (Alliance for Excellent Education, 2011). Dropping out of school not only limits one’s life-time opportunities, it also creates a social cost. Legislation and social programs have been implemented to keep kids in school and increase their employment opportunities (Campolieti, Fang, & Gunderson, 2010; Kuenzi, 2007). However, later life opportunities for high school dropouts remain meager, and many become welfare recipients (Alliance for Excellent Education, 2011; Danziger et al., 1999; Schwartz, 1995). Existing studies assume a homogenous population of high school dropouts; however, this population is likely heterogeneous (Brown, 1982; Cluck, Beaulieu, & Barfield, 1998; Waldinger & Lichter, 2003). However, one fact that cannot be overlooked is that not every high school dropout ends up in a dead-end job. Kusmin and Gibbs (2000) demonstrate that one fifth of initial entry jobs held by less-educated workers can lead to subsequent better
∗ Tel.: +82 10 8735 1184; fax: +82 2 3485 5140. E-mail address: [email protected]
paying jobs, implying that some dropouts acquire jobs that require at least a high school diploma or beyond; hence, some earn enough money to support themselves without public assistance. Kusmin and Gibbs (2000) identify initial jobs that move less-educated workers up occupation ladders into better paying jobs. However, few studies attempt to identify career trajectories of US high school dropouts, and it is important to identify the occupational trajectories of these less-educated workers if there is to be successful intervention strategies. Other dropouts located at the bottom of the occupational hierarchy require an occupational intervention that complements their low education levels to help them move into middle-level occupations. Moreover, finding the traits related to lower occupational career patterns better orients policies that create different opportunities for the status of dropouts. The aim of this study is, therefore, to identify how high school dropout career trajectories are distributed and what dropout characteristics contribute to the heterogeneous career trajectories. 2. Career mobility among high school dropouts Several studies demonstrate the later-life occupations held by dropouts in dead-end (Brown, 1982) and secondary jobs (Waldinger & Lichter, 2003). In dead-end jobs, work
0362-3319/$ – see front matter © 2013 Western Social Science Association. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.soscij.2013.03.005
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
experience increases wages by only a little (Connolly & Gottschalk, 2002); hence, dropouts in dead-end jobs are resigned to sub-standard career development. Studies on dead-end jobs emphasize that dropouts sort into lowstatus jobs according to little human capital, and their careers are likely stuck in lower occupational statuses if they do not enhance their skills and knowledge (Farkas, England, Vicknair, & Kilbourne, 1997; Mouw & Kalleberg, 2010). However, some jobs, even for dropouts, can be stepping stones to higher status occupations (Kusmin & Gibbs, 2000). One approach is work-first, which emphasizes that work experience accumulation may lead dropouts to advance up occupational ladders regardless of their entry level (Holcomb, Pavetti, Ratcliffe, & Riedinger, 1998). US high school dropouts have more frequent turnover and have less attachment to the labor market. For dropouts, high worker turnover can be improved with early employment intervention (Daguerre, 2007). 3. Factors leading to occupational mobility Studies show that occupational mobility differs by demographics, including gender, race, and parental background (Grusky, 2008). Moreover, women are disadvantaged more than men from lower occupational status due to occupational segregation (Sewell, Hauser, & Wolf, 1980), lower occupational mobility (Topel & Ward, 1992), and labor market discrimination (Maume, 1999). However, opportunities are changing with manufacturing off-shoring, which allows workers to change jobs from the manufacturing to the service industry (Schrammel, 1998). With this off-shoring trend, male dominated occupations have become scarce (Schrammel, 1998) and pushed male workers into lower status occupations compared with women, because women are more skilled with service sector skills. Minorities continue to be discriminated against (Grusky, 2008); there are ability differences (Herrnstein & Murray, 1994) and residential segregation persists (Wilson, 1996). Different distributions also exist for labor market information, and social capital differences between families may have an impact on dropout career development. Parental background through social capital on children’s occupational outcomes creates heterogeneity across dropouts and less-educated and single parents have fewer resources, which also creates unobserved variation across dropouts (Evans, Kelly, Sikora, & Treiman, 2005; Lin, Ensel, & Vaughn, 1981). Studies also demonstrate that occupational mobility differs by individual traits, such as human capital and self-esteem. First, efficiency differences and investment effectiveness by cognitive ability create different opportunities that are related to the occupational status of dropouts (Farkas et al., 1997). Workplace skill-biased technological change (SBTC) leads to increased demands for skilled workers, which accentuate the relationship between cognitive ability and wages (McCall, 2000). SBTC also favors worker computer use skills, which results in higher wage returns (Autor, Katz, & Krueger, 1998; Hotchkiss & Shiferaw, 2011). Moreover, the workplace digital divide is related to computer use at home (Pliskin, Levy, Heart, O’Flaherty, & O’Dea,
307
2006). Studies also demonstrate self-esteem as a labor market advantage, and high self-esteem workers are more likely to hold high status (Heckman & LaFontaine, 2006; Rosenberg & Owens, 2001). Workers with low self-esteem may take on a low status jobs (Gottfredson, 1981; McNatt & Judge, 2004). These unobserved characteristics have not been considered for US dropouts and their later life job opportunities. 4. Methods 4.1. Data Data from the National Longitudinal Study Youth 97 (NLSY 97) 1997 to 2008 are used to consider career trajectories of US high school dropouts. The NLSY 97 is a longitudinal survey of 8984 respondents aged 12–16 as of December 31, 1996 (U.S. Department of Labor, 2009). Respondents in NLSY 97 are interviewed each year for information about school-to-work transitions, including schooling and participation in the labor market. NLSY 97 consists of two samples: a cross-sectional sample that represents the U.S. population born between 1980 through 1984, and a supplemental sample of black and Hispanic populations born in the same years. Sample weights are applied to adjust for differences in the probability of selection due to minority oversampling. To investigate traits related to dropout occupational standing, we limit the sample to those who dropped out of high school by 2003. This is the expected graduation year if 12 year old respondents as of December 31, 1996 finished high school in 4 years. This decreases the number of observations from 8984 to 1813. Moreover, 438 respondents are excluded who enrolled in college after acquiring a GED by 2008, which is the year of the latest release of NLSY 97. This decreases the sample size to 1555. 4.2. Dependent variable 4.2.1. Occupational career Occupational standing over time is used as the dependent variable, which has information on “a degree of prestige or social standing” (Miech, Eaton, & Liang, 2003, p. 441). Ever since Duncan developed socioeconomic index (SEI), many scholars have updated his SEI. This study uses updated Duncan’s SEI measure by Nakao and Treas (1994). Occupations are those respondents held at the beginning of each year. 4.3. Independent variables 4.3.1. Years since high school dropout Years since high school dropout is used for the time scale variable, which is the baseline of the growth model in occupational standing. Years since high school dropout are a proxy for work experience. Due to missing data in work history, it is difficult to use the actual cumulative work experience in the growth model. This study uses Mincer’s approach (1974) for work experience, which is age minus years of schooling. Mincer’s approach measures the
308
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
potential work experience, which assumes a continuous work history. 4.3.2. Demographic variables This study considers five demographic variables: family structure, parental education, gender, language at home, and race. Family structure is classified into three groups: single- and two-parent households and other types of households, such as those headed by grandparents or children. Two-parent households are the reference group. Parental education is coded as a series of dummy variables: less than high school, 2nd year college, and 4th year college or above. High school is the reference for parental education. Home language other than English is dummy coded with the reference of English speaking. Race is dummy coded for black and Hispanic. White is the reference for race. 4.3.3. Computer at home Computer at home measures whether respondents had access to a computer at home during their high school years and is dummy-coded. No computer at home is the reference. 4.3.4. Cognitive ability Two cognitive ability variables are used: ArmedServices Vocational Aptitude Battery (ASVAB) and grade completed. ASVAB is the percentile scores measured at ages 13–15 for four tests measuring verbal and mathematical ability: arithmetic reasoning, mathematical knowledge, word knowledge, and paragraph comprehension. Ageadjusted ASVAB scores are used to adjust for a cohort difference at the time of measurement. Grade completed is the total years in school before dropping out. 4.3.5. Self-esteem Self-esteem in NLSY 97 is measured with the Rosenberg scales (Rosenberg, 1965). A mean index of the self-esteem scale, which is composed of the following four items: (1) I hardly ever expect things to go my way, (2) I rarely count on good things happening to me, (3) I am always optimistic about my figure, (4) in uncertain times, I usually expect the best. The items are scaled at four levels: 1 – strongly disagree, 2 – disagree, 3 – agree, and 4 – strongly agree. The first two negative items are coded for higher scores to reflect higher self-esteem (Cronbach’s ˛ = 0.68). 4.4. Data analyzes 4.4.1. Latent class growth analysis (LCGA) The latent class growth analysis (LCGA) is employed to classify a number of subgroups with heterogeneous trajectories as well as to identify the shape of subgroup trajectories, and the purpose of LCGA is to classify a population into an unknown set of latent classes (Duncan, Susan, Strycker, & Okut, 2002; Vaughn & Witko, 2012). Moreover, assuming that a single growth model specifies a population, standard latent growth analysis produces a single mean trajectory of all individuals in a sample (Jung & Wickrama, 2008). However, a single model may offset different trends
between heterogeneous groups (Nagin, 2005). LCGA models several trajectories of homogeneous workers within groups by classifying dropouts who show similar patterns of occupational trajectories and those who can be grouped together (Nagin, 1999). Applying LCGA allows for heterogeneous career trajectories of dropouts. LCGA analyzes continuous outcomes measured repeatedly over time. Trajectory classes in LCGA are defined as a set of individuals who follow a unique pattern (Luyckx, Schwartz, Goossens, Soenens, & Beyers, 2008). Thus, the growth parameters in LCGA, such as the intercept and slopes, can be different between groups. LCGA is a groupbased statistical technique and a model fit index classifies individuals into groups. LCGA is conducted using MPlus. 4.4.2. Model fit index To determine the number of latent trajectory classes, three different types of model fit indexes are used, such as Bayesian information criteria (BIC), Lo–Mendell–Rubin (LMR), and entropy (Jung & Wickrama, 2008). The model with the smallest BIC [BIC = (−2 log likelihood + p log(n)), where p is the number of parameters and n is the sample size] is preferred. LMR tests, which compare k class model and k − 1 class model, provide a model fitting better the data. Significant LMR value indicates that a model with k class is preferred to a model with k − 1 class. The closer the entropy value is to one, the better the model performs.1 4.4.3. Multinomial logistic regression After selecting the class number, a multinomial logistic regression model is estimated to identify variables predicting class membership. To identify how dropout traits are associated with occupational advancement, a multinomial logistic model is used on the dropout data. 4.4.4. Missing data Missing data is imputed for individual characteristics, such as cognitive ability and parental education using the MI command in Stata. 5. Results 5.1. Descriptive statistics Descriptive statistics are in the second column of Table 1. There are more male respondents (58%) than female respondents (42%). Fifty-eight percent of the respondents are white, while black, Hispanics, and other minorities represent 21%, 8%, and 13% of the sample, respectively. About 48%, 34%, and 27% are from two-parent, single-parents, and no-parents family, respectively. Furthermore, approximately 33% of parents have less than high school, 48% have only high school, 12% have 2nd year college, and 7% of the respondents have parents with 4 year
1 Since LCGA models the patterns of change in a dependent variable across multiple time intervals, it differs from an analytic technique for finding group differences common with analysis of variance or analyzing changes between two discrete time intervals, such as t-tests (Curran and Muthen, 1999).
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
309
Table 1 Descriptive statistics and distribution of three-trajectory class by individual traits. Variable
Overall
Men Black Hispanic Other-race Single-parent Other-parent Less than HS 2nd year college 4th year college Other language at home Computer at home Grades completed (M) ASVAB (M) Self-esteem (M)
.58 (.49) .21 (.41) .08 (.26) .13 (.33) .34 (.47) .27 (.44) .33 (.48) .12 (.34) .07 (.26) .17 (.39) .62 (.49) 9.91 (2.72) 31.32 (23.31) 2.73 (.34)
Three-trajectory class Dead-end
Stepping-stone
Advancing
F
.59 (.49)a .25 (.43)a .07 (.26) .12 (.33) .35 (.48)a .26(.44) .34 (.48) .11 (.33) .06 (.25) .17 (.39) .57 (.50)a 9.72 (1.22)a 29.62 (19.54)a 2.75 (.34)
.58 (.49)a .13 (.33)b .09 (.28) .14 (.35) .25 (.46)b .31 (.46) .31 (.47) .14 (.36) .08 (.29) .15 (.37) .70 (.47)b 10.21 (2.09)b 34.02 (19.92)b 2.72 (.34)
.38 (.49)b .17 (.38)a .08 (.28) .14 (.35) .45 (.49)a .28 (.49) .31 (.46) .11 (.37) .07 (.33) .18 (.37) .74 (.44)b 10.11 (1.16)b 35.61 (22.26)b 2.74 (.30)
** **
**
** + *
Source: The National Longitudinal Study Youth 97. Notes: Descriptive statistics with different subscripts differ. M stands for mean. + p < .07. * p < .05. ** p < .01.
college degrees. Seventeen percent of the sample speaks a language other than English at home, and 62% had a computer at home during high school. The average grade completed is 9.91. The average percentile score for ASVAB is 31.32, and the average score for self-esteem is 2.73.
5.2. Identifying the number of classes Model estimation begins with a normal latent growth model, and respondents have different time spans according to their cohort membership at time of dropout. A cohort-sequential design is used to estimate the growth model. To identify the time dimension variable, a likelihood test is conducted between linear and quadratic models. The difference in the likelihood statistics between the two models is significant [2 (56, 4) = .000, p < .001], indicating that a quadratic model better fits the data than a liner model. However, in the latter process of identifying a latent class, a quadratic factor is statistically insignificant. Thus, a linear model is specified for further analysis. An unconditional latent cluster model is then specified by constraining growth factor variance and covariance. This assumes the intercept and growth rates are zero. These constraints make individual growth trajectories homogenous within the same class (Jung & Wickrama, 2008). Model fit statistics determines the number of latent classes. Different models are tested by increasing the numbers of clusters. Table 2 presents model fit statistics for cluster models 1 through 4. To determine the number of classes, model fit indexes, such as BIC, entropy and LMR, are used. In applying the model fit index, Kusmin and Gibbs’ finding (2000) is used, which suggests that most high school dropouts would be stuck in dead-end jobs; however, a few overcome the low initial job placement associated with leaving high school without a degree. As a result, most dropouts fall into lower-stable trajectories, whereas a few reach increasing trajectories. The BIC index, indicating that higher values are a better fit, supports a four-class solution. However,
entropy supports the two-class solution indicating that higher values accurately classify individuals into latent classes based on the posterior probability classification. LMR demonstrates indicates that current class inclusion leads to a better fit, and supports the three-class solution. Between two- and three-class solutions, BIC and LMR favor the three-class solution over the two-class solution. The three-class model places individuals into each trajectory with 88%, 80%, and 90% accuracy. Fig. 1 displays fitted growth models for the three-class solution. Class 1 is individuals who began at lower occupation levels and remained in low status jobs. Class 1, is about 69% of the sample and depicts dead-end careers. Class 2 shows a separate career trajectory compared with class 1 and begins at a low occupation level, but increases occupational status over time. Class 2 represents stepping-stone
Fig. 1. Fitted growth lines of occupational status for three-trajectory classes. Source: The National Longitudinal Study Youth 97.
310
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
Table 2 Identifying the number of trajectory class. No. of classes
Model fit index
Posterior probability for 3-class model
BIC
Entropy
LMR
1
2
3
1 2 3 4
57,066 55,345 55,056 54,948
– .90 .72 .67
– .00 .05 .70
.883 .037 .004
.115 .803 .097
.001 .160 .902
Proportions of latent classes
.694 .231 .075
Source: The National Longitudinal Study Youth 97. Notes: Three-class model is selected. Analyzes are conducted using Mplus.
careers and is around 23% of the sample. Class 3 begins with a moderate occupation level and occupational careers increases over time. Class 3 is around 8% of the sample, represents advancing careers, and indicates that low educational qualifications do not influence occupational status. Thus, a majority of high school dropouts end up with deadend careers (69%), while about 31% of them experience career advancement over time. Each class’ estimated growth factor is presented in Table 3. Stepping-stone and advancing careers have increasing occupational growth trajectories, whereas dead-end careers and stepping-stone careers have similar initial occupational status. The intercept and slopes for the three classes demonstrate that for stepping-stone careers, both the intercept and slope are significant. Initial occupational status is 35.78, and the growth rate is 1.11. Stepping-stone careers of dropouts predict that workers hold onto jobs located at the median of the occupation scale for 10 years after high school dropout. For dead-end careers, only the intercept is significant and their initial occupational status is 33.47 with a growth rate of −0.08. The predicted occupational status for 10 year after dropout is 33.47; hence, dropouts in this career remain at the bottom of the occupational distribution. For advancing careers, initial occupational status is 50.52, and the growth rate is 1.15. The predicted value of occupational status 10 years after high school dropout is 62, and the predicted occupational status among different career types 10 years after dropout is considerable. 5.3. Identifying traits related to class trajectory membership Table 1 columns 3–6 present cross-tabulations of trajectory classes according to dropout traits. A one-way analysis of variance with adjustment for alpha levels demonstrates Table 3 Estimates for growth factors by trajectory class. Trajectory class
1. Dead-end careers 2. Stepping-stone careers 3. Advancing careers Total sample
significant class composition differences by gender, black, single parent, computer at home, grades completed in high school, and ASVAB. A multinomial logistic regression is estimated to examine traits related to class trajectory membership, and the reference category is dead-end career trajectory. Dead-end careers are used as the denominator, therefore, contrasts with other trajectory patterns that allow for testing for the traits related to dropouts’ occupational advancement.2 Traits that effect trajectory membership are gender, black, computer at home, and cognitive ability. Family structure, parental education, other language at home, and self-esteem are not significant. Table 4 includes multinomial logistic regression coefficients as logged odds, odds ratios, and confidence intervals. For male dropouts, the odds of being in dead-end careers are 0.40 times the odds for being in an advancing career, indicating that male dropouts are less likely of being in advancing careers. Men are more likely to be in a deadend career than an advancing career. The odds of blacks being in a stepping-stone career are 0.41 times that of a dead-end career, which means that among the cohort of dropouts that blacks are less likely to advance occupationally. There is, however, no difference in the composition of blacks in dead-end careers and advancing careers. If the initial occupational status of black workers is low, they stay in lower job tracks. However, if equipped with a computer at home, dropouts are more likely to be in stepping-stone careers or in advancing careers. The odds of being in stepping-stone and advancing careers for dropouts with a computer in the home during high school are 1.61 and 2.26 the odds for being in a dead-end career, which means they are 61% and 126% more likely of being in stepping-stone and career advancing careers later in life if they have a computer in the home during high school. Finally, the odds of being in a stepping-stone career with a one unit increase in cognitive ability are 1.01 times that of being in a dead-end career. Therefore, workers with greater cognitive abilities are more likely to be in stepping-stone than in dead-end careers.
Growth factor estimate Mean intercept
Mean linear slope
6. Discussion
33.47*** 35.78*** 50.52*** 35.41***
−.08 (.10) 1.11*** (.31) 1.15** (.54) .34*** (.07)
This study considers career trajectories of high school dropouts. Given dropout heterogeneity, the types of career
(.55) (1.13) (4.11) (.41)
Source: The National Longitudinal Study Youth 97. Notes: Analyzes are conducted using Mplus. ** p < .01. *** p < .001.
2 Multinomial logistic regression can be denoted by ln(Ps/Pd) and ln(Pa/Pd), where Ps, Pa, and Pd represent the probabilities of being in stepping-stone, advancing, and dead-end careers.
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
311
Table 4 Results of multinomial logistic model for trajectory class membership. Variable
ˇ (stepping-stonea )
eˇ (95% CI)
ˇ (advancinga )
eˇ (95% CI)
Men White (reference) Black Hispanic Other race Dual-parent (reference) Single-parent Other-parent High school (reference) Less than HS 2nd year college 4th year college Other language at home Computer at home Grades completed ASVAB Self-esteem Constant
−.11 (.13)
.89 (.69; 1.15)
−.74** (.21)
.48 (.32; .72)
−.51** (.18) .23 (.22) .13 (.21)
.59 (.42; .85) 1.26 (.03; .82) 1.14 (.04; .75)
−.19 (.30) .15 (.40) .04 (.34)
.83 (.46; 1.49) 1.16 (.53; 2.54) 1.04 (.53; 2.03)
−.25 (.16) .09 (.17)
.78 (.57; 1.07) 1.09 (.78; 1.53)
.39 (.26) .43 (.29)
1.48 (.89; 2.46) 1.54 (.87; 2.71)
.03 (.16) .19 (.22) .18 (.28) −.18 (.19) .51** (.14) .06 (.05) .01+ (.00) −.02 (.18) −1.95** (.75)
1.03 (.75; 1.41) 1.21 (.79; 1.86) 1.20 (.69; 2.07) .84 (.58; 1.21) 1.61 (1.15; 2.25) 1.06 (.96; 1.17) 1.01 (.99; 1.02) .98 (.69; 1.39) .14 (.03; .62)
−.18 (.27) −.30 (.37) .09 (.41) .43 (.28) .82** (.24) .06 (.06) .01 (.00) .24 (.43) −4.14** (1.24)
.84 (.49; 1.42) .74 (.36; 1.53) 1.09 (.49; 2.44) 1.54 (.89; 2.66) 2.26 (1.42; 3.63) 1.06 (.94; 1.19) 1.01 (1.01; 1.02) 1.27 (.55; 2.95) .02 (.00; .18)
Model fit
F(30, 6015) = 3.16, p < .001
Source: The National Longitudinal Study Youth 97. Notes: Analyzes are conducted using Mplus. a Reference is dead-end career. + p < .07. ** p < .01.
trajectories that exist for dropouts and their traits are related to trajectory membership. Using latent class growth analysis, three kinds of career trajectories are identified: dead-end career, stepping-stone career, and advancing career. While the majority of dropouts are in the dead-end careers, about 31% are in the process of escaping low-status jobs through accumulating work experience. Individual traits, such as gender, race, and cognitive ability, and having a computer in the home at early ages, are significantly related with different career trajectories. A majority of dropouts (69%) are trapped in low-status jobs and thus have little mobility. For these workers, acquiring work experience provides no benefit. About 23% of dropouts start out in low-status jobs; however, they escape mundane entry-level jobs and move into better jobs, and initial experience from low-status jobs becomes a stepping stone to a better position. About 8% of dropouts, despite their academic disadvantage, start at middle-status jobs and develop their careers. The traits of dropouts are related to their career trajectories. Despite institutional improvements for different ethnic groups, black dropouts are more likely end up in dead-end careers relative to their white counterparts. However, other minorities are not disadvantaged in their career trajectories. Second, a computer in the home during school is related to the likelihood of obtaining a stepping-stone or an advancing career. The benefit of having a computer at home prepares workers for the modern labor market, and the lack of computer skills may drive dropouts to more rudimentary jobs that do not offer career advancement. Third, for workers at low-level jobs, cognitive ability continues to shape their life opportunities. This may indicate that intelligent workers acquire required skills more easily and allow them to move out of dead-end jobs with greater work experience. Fourth, gender-based
occupational differentials for dropouts are reversed. Female workers are more likely to follow advancing careers than their male counterparts, which is consistent with prior studies that gender wage gaps narrow (Holzer & Hlavac, 2011). The advantage of being female among less-educated workers may be attributable to the change in an industrial mix, and a growing service sector but declining manufacturing sector. References Alliance for Excellent Education. (2011). The high cost of high school dropouts: What the nation pays for inadequate high schools. Retrieved from:. http://www.all4ed.org/files/HighCost.pdf Autor, D., Katz, L., & Krueger, A. (1998). Computing inequality: Have computers changed the labor market? The Quarterly Journal of Economics, 113(4), 1169–1213. Brown, C. (1982). Dead end jobs and youth unemployment. In R. Freeman, & D. Wise (Eds.), The youth labor market problem (pp. 427–452). Chicago: University of Chicago Press. Campolieti, M., Fang, T., & Gunderson, M. (2010). Labor market outcomes and skill acquisition of high-school dropouts. Journal of Labor Research, 31(1), 39–52. Cluck, R. E., Beaulieu, L. J., & Barfield, M. A. (1998). To the educated, the spoils: The relation of education to labor market experiences of young adults (ERIC Document Reproduction Service No. ED432418). Connolly, H., & Gottschalk, P. (2002). Job search with heterogeneous wage growth transitions to “better” and “worse” jobs. Retrieved from:. http://www.economics.neu.edu/papers/documents/03-003.pdf Curran, P. J., & Muthen, B. O. (1999). The application of latent curve analysis to testing developmental theories in intervention research. American Journal of Community Psychology, 27, 567–595. Daguerre, A. (2007). Active labor market policies and welfare reform. Basingstoke, Hants: Palgrave. Danziger, S. K., Corcoran, M., Danziger, S., Heflin, C., Kalil, A., & Levine, J. (1999). Barriers to the employment of welfare recipients. Retrieved from:. http://www.psc.isr.umich.edu/pubs/pdf/rr02-508.pdf Duncan, T. E., Susan, S. C., Strycker, L. A., & Okut, H. (2002). Growth mixture modeling of adolescent alcohol use data: Chapter addendum to An Introduction to Latent Variable Growth Curve Modeling: Concepts, Issues, and Applications. Eugene, OR: Oregon Research Institute.
312
K.-N. Kim / The Social Science Journal 50 (2013) 306–312
Evans, M., Kelly, J., Sikora, J., & Treiman, D. (2005). To the scholar go the spoils? The influence of parents’ scholarly culture on offspring’s occupational attainment in 31 countries. Paper presented at the annual meeting of International Sociological Association, Los Angeles, CA. Farkas, G., England, P., Vicknair, K., & Kilbourne, B. (1997). Cognitive skill, skill demands of jobs, and earnings among young European-American, African-American, and Mexican-American Workers. Social Forces, 75, 913–940. Gottfredson, L. S. (1981). Circumscription and compromise: A developmental theory of occupational aspirations. Journal of Counseling Psychology, 28, 545–579. Grusky, D. B. (2008). Social stratification: Class, race, and gender in sociological perspective (3rd ed.). Boulder, CO: Westview Press. Heckman, J. J., & LaFontaine, P. (2006). Bias corrected estimates of GED returns. Journal of Labor Economics, 24, 661–700. Herrnstein, R. J., & Murray, C. (1994). The bell curve: Intelligence and class structure in American life. New York, NY: The Free Press. Holcomb, P. A., Pavetti, L., Ratcliffe, C., & Riedinger, S. (1998). Building an employment focused welfare system: Work first and other work-oriented strategies in five states. Washington, DC: The Urban Institute. Holzer, H. J., & Hlavac, M. (2011). An uneven road and then a cliff: U.S. labor markets, 2000–2010. Retrieved from:. http://www.s4.brown. edu/us2010/projects/authors lm.htm Hotchkiss, J. L., & Shiferaw, M. (2011). Decomposing the education wage gap: Everything but the kitchen sinks. Federal Reserve Bank of St. Louis Review, July, 243–272. Jung, T., & Wickrama, K. (2008). An introduction to latent class growth analysis and growth mixture modeling. Social and Personality Psychology Compass, 2, 302–317. Kuenzi, J. (2007). High school graduation, completion, and dropouts federal policy, programs, and issues (RL-33963). Retrieved from:. http://www.lieberman.senate.gov/assets/pdf/crs/highschool.pdf Kusmin, L., & Gibbs, R. (2000). Less-educated workers face limited opportunities to move up to good jobs. Rural America, 15(3), 33. Lin, N., Ensel, W. M., & Vaughn, J. C. (1981). Social resources and strength of ties: Structural factors in occupational status attainment. American Sociological Review, 46, 393–405. Luyckx, K., Schwartz, S. J., Goossens, L., Soenens, B., & Beyers, W. (2008). Developmental typologies of identity formation and adjustment in female emerging adults: A latent class growth analysis approach. Journal of Research on Adolescence, 18, 595–619. Maume, D. J., Jr. (1999). Glass ceilings and glass escalators: Occupational segregation a race and sex differences in managerial promotions. Work and Occupations, 26(4), 483–509. McCall, L. (2000). Explaining levels of within-group wage inequality in U.S. labor market. Demography, 37, 415–430. McNatt, D. B., & Judge, T. A. (2004). Boundary conditions of the Galatea effect: A field experiment and constructive replication. Academy of Management Journal, 47, 550–565.
Miech, R. A., Eaton, W., & Liang, K. Y. (2003). Occupational stratification over the life course: A comparison of occupational trajectories across race and gender during the 1980s and 1990s. Work and Occupations, 30, 440–473. Mincer, J. (1974). Schooling, earnings, and experience. New York: Columbia University Press. Mouw, T., & Kalleberg, A. L. (2010). Stepping stone versus dead end jobs: Occupational pathways out of working poverty in the United States, 1996–2007. Retrieved from:. http://www.unc. edu/∼tedmouw/papers/stepping%20stone%20versus%20dead%20end %20jobs%2008102010.pdf Nagin, D. (1999). Analyzing developmental trajectories: A semiparametric, group-based approach. Psychological Methods, 4(2), 139–157. Nagin, D. S. (2005). Group-based modeling of development. Cambridge, MA: Harvard University Press. Nakao, K., & Treas, J. (1994). Updating occupational prestige and socioeconomic scores: How the new measures measure up. In P. Marsden (Ed.), Sociological methodology (pp. 1–72). Washington, DC: American Sociological Association. Pliskin, N., Levy, M., Heart, T., O’Flaherty, B., & O’Dea, P. (2006). The corporate digital divide between smaller and larger firms. In E. Trauth, D. Howcroft, T. Butler, B. itzgerald, & J. DeGross (Eds.), Social inclusion: Societal and organizational implications for information systems (pp. 413–417). Boston, MA: Springer. Rosenberg, M. (1965). Society and the adolescent self-image. Princeton: Princeton University Press. Rosenberg, M., & Owens, T. (2001). Low self-esteem people: A collective portrait. In T. J. Owens, S. Stryker, & N. Goodman (Eds.), Extending self-esteem theory and research (pp. 400–436). New York: Cambridge University Press. Schrammel, K. (1998). Two generations of young adults: Comparing the labor market success of young adults from two generations. Monthly Labor Review, February, 3–9. Schwartz, W. (1995). School dropouts: New information about an old problem. Retrieved from:. http://www.ericdigests.org/ 1996-2/dropouts.html Sewell, W. H., Hauser, R., & Wolf, W. (1980). Sex, schooling, and occupational status. American Journal of Sociology, 863, 551–583. Topel, R. H., & Ward, M. P. (1992). Job mobility and the careers of young men. The Quarterly Journal of Economics, May, 439–479. U.S. Department of Labor. (2009). NLSY97 user’s guide. Retrieved from:. http://www.nlsinfo.org/nlsy97/nlsdocs/nlsy97/maintoc.html Vaughn, M., & Witko, C. (2012). Does the amount of school choice matter for student engagement? The Social Science Journal, http://dx.doi.org/10.1016/j.bbr.2011.03.031 [Advance online publication] Waldinger, R., & Lichter, M. (2003). How the other half works: Immigration and the social organization of labor. Los Angeles, CA: University of California Press. Wilson, W. (1996). When work disappears: The world of the new urban poor. New York: Alfred A. Knopf.