The Benefits of Attending Community College a Review of the Evidence

• Journal List
• HHS Author Manuscripts
• PMC4375965

Sociol Sci. Author manuscript; bachelor in PMC 2015 Oct 27.

Published in final edited form every bit:

PMCID: PMC4375965

NIHMSID: NIHMS670771

The Community College Outcome Revisited: The Importance of Attention to Heterogeneity and Circuitous Counterfactuals^*

Jennie East. Brand

University of California–Los Angeles

Fabian T. Pfeffer

University of Michigan

Sara Goldrick-Rab

University of Wisconsin–Madison

Abstract

Customs colleges are controversial educational institutions, ofttimes said to simultaneously expand higher opportunities and diminish baccalaureate attainment. We appraise the seemingly contradictory functions of community colleges by attending to effect heterogeneity and to alternative counterfactual conditions. Using data on postsecondary outcomes of high schoolhouse graduates of Chicago Public Schools, we find that enrolling at a community college penalizes more advantaged students who otherwise would have attended four-yr colleges, particularly highly selective schools; however, these students stand for a relatively modest portion of the community college population, and these estimates are virtually certainly biased. On the other manus, enrolling at a customs higher has a modest positive event on available'southward degree completion for disadvantaged students who otherwise would non have attended college; these students correspond the majority of community college goers. We conclude that discussions among scholars, policymakers, and practitioners should motion across because the pros and cons of community college attendance for students in general to attention to the implications of community college attendance for targeted groups of students.

Keywords: community college, bachelor's degree, heterogeneity, Chicago Public Schools

Community colleges are amongst the most controversial educational institutions (Goldrick-Rab 2010). They are alternatively depicted every bit creating accessible, affordable, and expanded opportunities for postsecondary education (e.g., Cohen and Brawer 1982; Shaw, Goldrick-Rab, Mazzeo, and Jacobs 2006; Shavit, Arum, and Gamoran 2007), or as steering less advantaged youth away from selective colleges and universities (e.one thousand., Brint and Karabel 1989; Clark 1960; Karabel 1972). This combination of manifestly countervailing functions led Dougherty (1994) to draw the public ii-yr institution as a "contradictory college." Policymakers are increasingly concerned about the outcomes associated with attention community college as enrollment in the sector continues to abound, and every bit more than than 80 percent of entering students say they desire to earn a available's degree, only only well-nigh 12 percent complete that degree within half-dozen years (Century Foundation 2013). These figures, coupled with high levels of socioeconomic segregation within higher education, suggest that community college attendance may play an of import role in the caste to which American higher education perpetuates, or even exacerbates, social inequality.

At the eye of the debate is the and then-called "community college effect," an average approximate that is interpreted equally increasing or decreasing inequality in educational outcomes depending upon whether information technology is negative or positive, respectively. To the extent that customs colleges promote social mobility, individuals who practise not attend them are left farther backside. But if community colleges instead diminish opportunities for bachelor's degree completion by cartoon students away from baccalaureate-granting colleges, gaps in educational attainment may exist exacerbated. Over the final several decades, dozens of empirical studies take estimated both positive democratizing effects of community college as well as negative diversionary effects (e.g., Alba and Lavin 1981; Alfonso 2006; Brint and Karabel 1989; Clark 1960; Doyle 2009; Dougherty 1994; Grubb 1991; Leigh and Gill 2003; Long and Kurlaender 2009; Melguizo, Kienzl, and Alfonso 2011; Reynolds 2012; Reynolds and DesJardins 2009; Rouse 1995; Sandy, Gonzalez, and Hilmer 2006; Whitaker and Pascarella 1994). Yet current public conversations about community college often focus exclusively on the negative effects, with policymakers and practitioners warning about the penalization accruing to students seeking available'south caste who enter public two-year schools. In fact, in Chicago Public Schools, the site of this study, some schools have began to actively discourage students from attending community college, urging them to find a ameliorate college match (Roderick et al. 2011; Bowen et al. 2009).

In this paper, we contend that an accurate characterization of community colleges depends upon a clearer agreement of the extent to which the effects are heterogeneous—simultaneously advantaging some students while disadvantaging others. Given the vast and growing compositional heterogeneity among undergraduate students, it is unlikely that attention community college affects all students in the aforementioned way. Past systematically attending to the probable alternative paths community college students would have otherwise followed, and to community college upshot heterogeneity, we can improve interpret the outcomes associated with community college attendance. This increases the accurateness with which we describe mechanisms of social stratification and affects the positioning of community colleges within the educational policy landscape, where they are increasingly critiqued for what appear to be poor outcomes (Goldrick-Rab 2010).

We assess multiple treatment effects of community college attendance on bachelor's degree completion using rich longitudinal survey and administrative data on the postsecondary trajectories of students graduating from Chicago Public Schools. Although community colleges serve multiple functions, including technical grooming, remediation, and enrichment, implicating many potential outcomes of interest, arguably a primal function is providing an affordable and accessible route to a four-year degree. Thus, examining effects on available'due south degrees has been the focus of most contend and report. We test the hypothesis that the boilerplate democratizing and diversionary effects of community college attendance are in fact different effects for different students. We consider a range of counterfactuals for individuals who practice non attend community colleges and empirically demonstrate how the relevant alternatives to customs college attendance apply to subpopulations of students with different propensities to attend.

We find a modest positive (democratizing) consequence of community college attendance relative to students who do non attend postsecondary schooling within 1 twelvemonth of loftier school completion. These students mostly have a loftier propensity to attend customs college and represent the majority of the community higher population. Our results point that the purported penalty to attention customs higher may be overstated, since it accrues only to a modest subpopulation of students who would have otherwise attended selective, and especially highly selective, four-year colleges. We conclude that discussions amidst pedagogy and stratification scholars should motion beyond considering the pros and cons of community higher attendance for students in full general to attention to the implications of community college omnipresence for specific groups of students with differing propensities to attend.

Groundwork and Significance

The community college is a key contributor to the diversity of American higher education (Shavit, Arum, and Gamoran 2007). Public ii-year colleges absorbed much of the expansion in postsecondary enrollment that occurred in the mid-twentieth century, such that more than than 40 percent of all undergraduate students in the United States currently attend community college (American Association of Community Colleges 2011). Community colleges are tasked with maintaining easy admission to a college education while too providing a gateway to educational attainment and other socioeconomic opportunities (Cohen and Brawer 1982; Goldrick-Rab 2010). While often praised for remaining more affordable than other postsecondary options and offering a "second chance" at educational attainment (Rouse 1995), the community college has also been steadily attacked for low rates of available's degree completion among the population information technology serves. Some have suggested that community colleges further socioeconomic disparities in education (Grubb 1991; Brint and Karabel 1989), and that students, specially those seeking bachelor's degrees, are best advised to avoid community college attendance entirely (Guess 2008).

The main explanation for these seemingly disparate judgments is that analysts accept focused on dissimilar functions of the community college. On one manus, community colleges exist to provide some postsecondary education; on the other mitt, they are also expected exist an affordable and accessible gateway that facilitates access to baccalaureate-granting institutions via transfer. They appear to fulfill the first function fairly well and the 2nd not as well (Belfield and Bailey 2011; Goldrick-Rab 2010; Grubb 1991; Leigh and Gill 2003; Roska 2009; Roska and Keith 2008). Indeed, in an endeavor to focus on one function of community higher—i.e., access to baccalaureate degrees—some analysts restrict their samples to students with the stated intent to attain a bachelor's degree (Alfonso 2006; Doyle 2009; Leigh and Gill 2003; Long and Kurlaender 2009; Whitaker and Pascarella 1994). This arroyo treats educational expectations, which are known to be malleable and fluctuating (Morgan 2005; Reynolds et al. 2006), as static and decisive. Limiting variation amidst students may also truncate the range of estimated effects of attendance. ¹

There are further methodological and theoretical considerations that may play into the seemingly incompatible interpretations that by research accords to the effects of community college attendance. The estimation of the community college effect nigh commonly estimated by analysts is complicated if at that place is effect heterogeneity (Brand 2010; Brand and Simon Thomas 2013; Brand and Xie 2010; Morgan and Winship 2014; Xie, Brand, and Jann 2012). Community college attendance may yield positive effects for some subpopulations and negative furnishings for others. First, the estimated event of community college should differ according to the assumed counterfactual educational choice, whether it exist no postsecondary education or attendance at a non-selective or selective four-year college post-obit high school. This distinction underscores prior discussions of the divergent functions of the community higher, but information technology is more complex than that, as it requires precisely identifying how choice sets differ across the population. Community college attendance may increase admission to educational attainment amid disadvantaged students relative to their most likely counterfactual—no immediate college attendance (Roderick, Coca, and Nagaoka 2011; Rouse 1995; Sandy, Gonzalez, and Hilmer 2006). That is, if a large segment of the community college population would otherwise take no firsthand postsecondary pedagogy rather than attend a four-year college, and so scholars overstate the penalty to community college attendance past comparing community college students only to four-year college goers. Merely community higher attendance could simultaneously decrease bachelor's degree completion amongst advantaged students, whose probable counterfactual would be postsecondary education at a four-year college. Among community higher goers, the size of the disadvantaged population is likely larger than the size of the advantaged. Moreover, the majority of customs college goers who could have otherwise attended a four-yr higher would have attended a non-selective four-year institution. Colleges of dissimilar levels of selectivity present disparate opportunities for students, particularly amid more disadvantaged students characteristic of customs college goers (Alon and Tienda 2005; Make and Halaby 2006; Dale and Krueger 2011). Thus, studies analyzing community college effects only among college goers (e.g., Doyle 2009; Long and Kurlaender 2009; Reynolds and DesJardins 2009; Whitaker and Pascarella 1994) set aside the demonstrably relevant counterfactual of no higher attendance, while others amass institutional types and mask the variable effects of different kinds of colleges (eastward.g., Alfonso 2006; Doyle 2009; Kane and Rouse 1995; Leigh and Gill 2003; Rouse 1995; Sandy, Gonzalez, and Hilmer 2006). It is notable given the diversion versus democratization debate in the literature how few studies simultaneously consider both alternatives.

Contempo research has attended to the potential outcomes associated with community college attendance and has adopted a propensity score framework to estimate furnishings (Doyle 2009; Long and Kurlaender 2009; Kalogrides and Grodsky 2011; Melguizo, Kienzl, and Alfonso 2011; Reynolds 2012; Reynolds and DesJardins 2009), merely this enquiry does not attend to the possibility that the estimated effect may differ across subpopulations. As in the vast majority of such studies, these approximate average treatment effects and assume away effect heterogeneity. Long and Kurlaender (2009) and Rouse (1995) employ instrumental variable (IV) models to estimate community higher effects, where distance to college is the instrument. ⁱⁱ They find smaller community college penalties using Iv models relative to OLS regression or propensity score models and suggest that this is the effect of unobserved heterogeneity. However, if there is effect heterogeneity, then IV estimates should be interpreted as local average treatment effects (LATE) that pertain to the population induced to attend community higher by the altitude and non to the total population of community college goers. Neither Long and Kurlaender (2009) nor Rouse (1995) interpret estimated effects as heterogeneous, pertaining to a subpopulation of community higher students defined according to option into handling. ^three Yet relating differential effects of community college attendance to the probability that students attend customs college yields important insights about how educational resource are distributed in order and the potential touch on of increasing or decreasing the population of community higher attendees (Brand and Xie 2010; Heckman et al. 2006).

Analytic Methods

For individual i, the event of community college is divers as the difference betwixt the potential outcome (in this case, bachelor's degree completion) in the community college state (i.e., the treated country, d=one) and the non-community higher state (i.e., the control state, d=0) (Morgan and Winship 2014):

Thus we ask whether students who started at a community college within a year of high schoolhouse graduation (d=1) take different outcomes than they otherwise would have had if they had not begun their postsecondary career by enrolling in a community college (d=0). Information technology is, of course, incommunicable to observe both outcomes for the same individual. If unobserved characteristics bear upon decisions to attend customs college and these characteristics are also correlated with eventual available'south degree completion, then the estimated effects of community higher will be biased. The selection on observables supposition tin can never be verified and should not be taken as true in practice for observational data; its plausibility depends upon the population under report and the availability of observed covariates. Measurement of meaningful confounders renders ignorability tentatively more than plausible, though still non necessarily truthful. All the same, such analyses offer the most the data can tell us without additional unverifiable assumptions, such as those imposed by an Iv approach. Recent studies of the community college upshot have recognized the challenges inherent in establishing the causal effects of community higher attendance with observational data (Doyle 2009; Long and Kurlaender 2009; Reynolds 2012; Rouse 1995).

Nosotros decompose the baseline counterfactual, no customs college omnipresence, into a multistate treatment condition, which entails a series of option equations comparing customs college attendance to control states defined past educational alternatives inside a year of high school graduation: (1) no postsecondary schooling; (2) omnipresence at a non-selective iv-year college; (3) attendance at a selective four-twelvemonth college; and (4) attendance at a highly selective four-year college. ⁴ Each of these treatment categories themselves represent complex treatments that could be further decomposed according to attendance patterns at later points in time. The no postsecondary schooling (within ane yr of high school completion) category includes individuals who never attended, as well as those who later went on to attend customs and iv-year colleges. Likewise, the diverse attending categories include students who start out at four-year colleges and who subsequently attend a variety of unlike colleges over their postsecondary career, including community colleges (i.e., "reverse transfer" students, see Goldrick-Rab and Pfeffer 2009). ⁵

Nosotros estimate a series of binary logit models for the probability of selection into our multistate treatment. Binary logit (or probit) equations are well developed in the matching literature, enable simple tests for common support and balance of covariates, and do not impose the independence of irrelevant alternatives (IIA) assumption required for a multinomial model. Although the multinomial approach benefits from formulating the complete gear up of alternatives in one model, derived conditional probabilities are not interdependent in binary models. As misspecification of 1 choice equation yields misspecification of all the conditional probabilities in the multinomial model, binary choice series estimation is potentially more robust than the multinomial approach (Lechner 2001).

Nosotros brainstorm past estimating simple bivariate associations, or unmatched mean differences, for the treatment-control states. We then estimate effects using propensity score matching, where individuals are matched according to their propensity for community higher attendance relative to each alternative (Morgan and Harding 2006). The main reward of matching compared to conventional regression models is conceptual. The weather nether which valid causal inference tin can be had is a primal focus in matching routines, including precisely defining the counterfactual conditions and assessing covariate residue between treated and untreated cases. We estimate propensity scores with a logit regression predicting the propensity of going to community higher of the following form:

P i = p ( d i = one | Ten ) = ln d i 1 − d i = ( ∑ k = 0 Chiliad β k Ten ik )

(2)

where P is the propensity score; d_i indicates whether private i (i = 1, …, n) attends community college or each of the four alternatives; and X represents a vector of observed pre-treatment covariates, described in more detail below. These propensity scores correspond estimates of individual likelihoods of attention community college relative to each culling. The community college effect is the difference in bachelor's degree completion between students with comparable propensities.

We tin ascertain treatment effects over several population subsets; we estimate the average treatment outcome on the treated (TT):

E(δ|d = 1) =E(y ^d=1 −y ^d=0|d = 1)

(3)

All matching estimators of the TT take the following general grade:

TT ^ = 1 / n 1 ∑ i n i ( y i , d = i − ∑ i ( j ) iJ w i ( j ) y i ( j ) , d = 0 ) ,

(4)

where n _i is the number of treatment cases; i is the alphabetize over handling cases; i(j) is the index over untreated cases for treated example i (i(j)=one,…i(J); and westward_{i ( j )} is the scaled weight (with sum of one) that measures the relative importance of each untreated instance. ⁶ Scholars have non reached a consensus as to which matching estimator performs best in each application, although nearest neighbor (with replacement) and kernel matching, which we use here, perform well in simulations (Morgan and Harding 2006; Morgan and Winship 2014). Morgan and Winship (2014) and Morgan and Harding (2006) advise researchers to examine multiple estimates of the same treatment upshot to establish a degree of robustness for the results.

In auxiliary analyses, we draw how the various estimated effects of community college attendance stand for to the estimated propensity for community higher attendance. Past revealing how effects differ among subpopulations defined co-ordinate to their selection into handling, we shed light on a central sociological question about the distribution of private opportunities. Another reward is the heightened recognition of potential violations of the selection on observables assumption across the population distribution (Brand and Simon Thomas 2013). That is, one estimation of variation in furnishings involves differential selection mechanisms on unobserved variables. We consider variation in community college furnishings by the propensity for customs higher attendance using a nonparametric method, the "smoothing-differencing" method (SD). SD consists of the following three steps (Xie, Brand, and Jann 2012): (one) guess propensity scores for each unit; (2) fit separate nonparametric regressions of the dependent variable on the propensity scores for the treated and untreated groups by local polynomial smoothing (degree ane, bandwidth 0.2); and (three) take the divergence in the nonparametric curves between the treated and the untreated to obtain the design of treatment event heterogeneity as a function of the propensity score. The SD method allows for heterogeneous treatment effects as a continuous function of the propensity score that we then chronicle to the effects estimated across the alternative counterfactual conditions.

Data and Sample

Decisions about attending community colleges are substantially local ones; very few students travel far from home to attend. The same is generally true for students attention non-selective, public four-year institutions (Goldrick-Rab 2010; Turley 2009). In big national samples with wide variation among students and colleges, this local concentration can lead to confusion betwixt treatment heterogeneity and treatment issue heterogeneity (i.east., dissimilar effects of attending community colleges with unlike characteristics versus dissimilar effects of community colleges for different students, respectively). Thus, examining heterogeneous furnishings of community colleges with national or country samples complicates the interpretation of effects relative to considering how attending a specific community college (or set up of colleges) exerts heterogeneous furnishings on the students it aims to serve. Heckman, Ichimura, and Todd (1997) emphasize the importance of comparison treatment and command groups in the same social and economic environment to minimize bias, a consideration that national samples clearly do not run across. With an eye toward addressing these issues, we guess effects of attending the Chicago Metropolis Colleges for the graduates of Chicago Public Schools on bachelor's caste completion. ^seven The tradeoff is, of course, that nosotros take express ability to generalize estimates of furnishings beyond Chicago and to students who attend private high schools.

We focus on Chicago considering information technology is amidst a scattering of urban schoolhouse districts that has for many years followed the trajectories of their graduates and collected data on students' groundwork characteristics and schooling. Chicago also represents the nation's fourth largest school district. Nearly half of students in Chicago Public Schools (CPS) enroll in higher within i year of loftier school, and half of those students (24 percentage of the sample) enter a higher granting available's degrees. More than than half of all CPS college goers enroll in x in-state colleges, most of them located within Chicago city limits. Among those attention iv-year colleges, nigh are enrolled at schools with graduation rates well beneath the national norm (Roderick, Coca, and Nagaoka 2011). ^eight The bachelor's caste completion rate within vi years of high school is about 11 percent, low by most any standard. This is not entirely surprising, given the students' relatively poor academic qualifications and high level of socioeconomic disadvantage (Roderick at al. 2008).

Data on students' pre-higher characteristics come from CPS and the surveys conducted by the Consortium for Chicago Schoolhouse Research (CCSR). We apply a wide range of measures affecting college choice. These include:

1. demographic characteristics (sex activity, race and ethnicity, citizenship, generational status)

2. social background characteristics (family construction, mothers education, Demography tract social status according to occupation and education, Census tract unemployment and poverty, and Census tract homeowner tenancy) ⁹

3. high school bookish accomplishment (cumulative course bespeak boilerplate, number of honors courses, number of AP courses, number of absences, placement in special education)

4. educational resource (number of educational resources at home, parental advice, parental interest);

5. educational aspirations and expectations (college aspirations, higher expectations, and parental expectations for higher)

6. high school characteristics (percentage of students who keep to a four-year college)

With the notable absence of measures of cerebral ability and straight measures of family income (we take merely indirect indicators through neighborhood characteristics), ours is a more than extensive prepare of covariates than typically employed in studies of community college choice. For example, due to limitations in the Ohio administrative source they employed, Long and Kurlaender (2009) only conditioned on a subset of potentially appreciable characteristics, and but for a select group of students (i.due east. those who had taken the Human action and aspired to consummate a bachelor's degree). Since the ACT is not required for admission to customs college, theirs is an particularly selective sample of students that excludes students who wanted to earn a bachelor's degree simply did not adequately prepare (a mutual phenomenon in Chicago; encounter Roderick et al. (2008)) or did non state while in high school an aspiration to earn that degree.

We estimate community higher effects for the class of 2001 cohort of CPS graduates. Most analyses estimating customs college effects include samples of much older cohorts (Goldrick-Rab 2010). We utilize a more contempo cohort of graduates who withal graduated from high school prior to contemporary reform efforts to increment higher omnipresence in Chicago. We estimate effects on higher completion within half dozen years of high school graduation. Six years is 150 percentage of the time typically needed to consummate a four-year degree as a total-time student and is commonly used in the higher education literature to calculate graduation rates. Seventy-vi pct of bachelor's degree recipients nationally consummate their caste within six years (National Center for Education Statistics 2009). Still, students who filibuster higher entry and those who enroll function-time may not reasonably consummate their degrees inside six years, and thus the customs college penalisation is likely overstated past applying a six-year completion restriction. ¹⁰

We condition our initial sample of CPS graduates (N=fourteen,322) on information availability; specifically, students must be included in the National Student Clearinghouse information (N=13,966) and have responded to at least one of the surveys administered past CPS in grades ix and 11 (N=9,533). Nosotros also exclude a small number of students who attended private 2-year colleges (46 students, or 0.iii percentage), and students attending four-year institutions of unknown selectivity (247 cases, or 1.7 percent). ¹¹ Missing information on survey measures is imputed using all other variables in our models. We present results based on one imputed dataset, merely note that the prediction results for the propensity score are stable to using x imputed datasets (results available from the authors upon request). ¹² Departing from some past research (e.g., Alfonso 2006; Doyle 2009; Leigh and Gill 2003; Long and Kurlaender 2009; Whitaker and Pascarella 1994) nosotros do not condition the sample on the stated aspiration to earn a bachelor's caste, for the reasons we draw in a higher place. ^xiii Instead, we advance comparability between treated and control groups by examining a local setting, workout on a rich gear up of exogenous covariates, restricting our analyses to regions of mutual back up (i.east., no pregnant covariate or propensity score differences between treated and command groups), and estimating effects to specific subpopulations based on culling counterfactual conditions.

Results

Descriptive Statistics

Higher-bound students face up many options in Chicago, including more than than fifty four-year colleges and universities (both selective and non-selective), and a organisation of seven community colleges. From Tabular array i, we detect that characteristics of community higher students are most similar to individuals who do not attend postsecondary education immediately post-obit loftier schoolhouse. Some variables advise that students who do non go on their schooling are more disadvantaged than community college goers; i.e., students who do non go on schooling are more probable to exist racial and ethnic minorities, live in more disadvantaged communities during high school, and have lower educational aspirations and expectations. Still, with the exception of race, the differences between customs college goers and non-college goers are smaller than the differences between community college goers and four-year college goers. Compared to iv-year college goers, community college students have less educated mothers, more than often live in not-intact families, reside in substantially more disadvantaged parts of the urban center, accept lower educational aspirations and expectations, and have lower academic accomplishment in loftier school (i.eastward., reach lower GPAs, take fewer honors courses and AP classes, and are absent more oftentimes). Equally expected, these differences are largest betwixt community college goers and highly selective four-year college goers.

Table one

Descriptive Statistics by Educational Path: Chicago Public Schools Sample

	Full Sample	Customs College	No Community College
			No Immediate College	Non-Selective 4-Yr Higher	Selective 4-Yr College	Highly Selective iv-Year Higher
Demographic Characteristics
Female person	0.591	0.609	0.553	0.622	0.657	0.638
White	0.126	0.131	0.089	0.164	0.123	0.231
Black	0.519	0.538	0.528	0.468	0.731	0.329
Hispanic	0.297	0.275	0.353	0.278	0.131	0.259
Other race	0.059	0.056	0.030	0.091	0.016	0.182
U.S. born	0.814	0.843	0.820	0.765	0.902	0.715
Second generation	0.396	0.406	0.380	0.442	0.189	0.571
Social Background Characteristics
Intact family	0.438	0.407	0.411	0.472	0.378	0.606
Mother college graduate	0.218	0.195	0.167	0.231	0.300	0.360
Neighborhood social status	-0.235 [0.799]	-0.265 [0.759]	-0.357 [0.758]	-0.074 [0.788]	-0.025 [0.788]	-0.010 [0.911]
Neighborhood non-poor	0.225 [0.799]	0.219 [0.826]	0.318 [0.764]	0.105 [0.829]	0.273 [0.819]	-0.070 [0.774]
Neighborhood homeowner	11.402 [4.199]	11.513 [4.244]	eleven.150 [four.034]	11.133 [4.131]	12.747 [4.733]	11.276 [iv.126]
High School Bookish Accomplishment
Cumulative GPA	ii.497 [0.821]	2.222 [0.655]	ii.225 [0.715]	two.646 [0.688]	2.792 [0.660]	3.529 [0.609]
Honors courses	0.792 [1.330]	0.382 [0.909]	0.460 [i.025]	0.845 [one.322]	ane.188 [1.484]	2.205 [ane.595]
AP credits	0.145 [0.495]	0.044 [0.253]	0.078 [0.362]	0.105 [0.408]	0.148 [0.469]	0.539 [0.879]
Absences	6.323 [half dozen.393]	6.939 [6.605]	vii.439 [six.992]	5.342 [5.152]	iv.846 [4.556]	iii.176 [iii.857]
Special education	0.083 [0.276]	0.095 [0.294]	0.116 [0.320]	0.068 [0.251]	0.020 [0.140]	0.005 [0.0728]
Educational Resources
No. of educ. resource at dwelling	five.298 [3.038]	four.928 [3.004]	4.809 [3.012]	v.553 [ii.949]	6.317 [ii.889]	6.630 [2.723]
Parental advice	four.459 [1.874]	4.382 [i.928]	four.273 [ane.941]	4.713 [1.719]	iv.766 [1.733]	4.850 [1.626]
Parental involvement	6.596 [2.690]	half-dozen.532 [two.787]	half-dozen.487 [2.754]	half dozen.864 [2.520]	half-dozen.822 [two.562]	6.752 [2.476]
Aspirations and Expectations
College aspirations	0.446	0.407	0.323	0.527	0.626	0.752
Higher expectations	0.368	0.324	0.252	0.391	0.539	0.696
College expectations parents	0.934	0.950	0.905	0.955	0.975	0.970
High School Characteristics
College-going charge per unit	0.476 [0.074]	0.481 [0.073]	0.471 [0.076]	0.484 [0.067]	0.496 [0.069]	0.463 [0.072]
Result
Bachelor'due south degree completion	0.114	0.028	0.016	0.107	0.174	0.541
North	9,533	1,772	4,704	740	ane,003	i,314

Higher choice is associated with the likelihood of bachelor degree completion for CPS graduates. As described in Tabular array 1, eleven.iv percentage (N=1,089) of graduates from the CPS grade of 2001 earned a available'southward degree by 2007. This figure includes ane.six percent (Northward=74) of students who did non nourish postsecondary school within a year of their high school graduation, ii.8 percent (Northward=fifty) of students who started at a community college, 10.seven percent (Due north=79) of students who attended a nonselective iv-year college, 17.4 per centum (Due north=175) of students who attended a selective 4-year college, and 54.i percent (N=711) of students who attended a highly selective iv-year college. Every bit we note above, this is a sample marked by high socioeconomic and academic disadvantage, and thus college completion rates are quite low relative to national averages.

Matching Analyses of Multistate Treatment Effects

To judge effects of community college attendance on bachelor's degree completion, we match community college goers to non-community higher goers who have like propensities to nourish. ^xiv Nosotros get-go guess the propensity of a student to nourish customs higher within one twelvemonth of completing loftier school relative to: (1) not attending postsecondary education; (2) attending a non-selective 4-year higher; (3) attending a selective four-year college; and (iv) attending a highly selective four-twelvemonth college. The results, reported in Table 2, advise that when comparing community college students to those who did non immediately pursue postsecondary schooling, high educational aspirations and high parental educational expectations predict customs higher attendance; by contrast, low educational aspirations, poor bookish preparation, and family disadvantage are significant predictors of community college omnipresence relative to those who attended non-selective four-year colleges. When comparing community college students to those who attended selective four-yr colleges, the onetime are more probable to be significantly disadvantaged with respect to family background, high school academic preparation, educational resource, and educational aspirations and expectations of students and parents. Relative disadvantages are larger withal when comparing community college students to those who attended highly selective 4-year colleges.

Tabular array 2

Logistic Regression Estimates for Models Predicting Community College Attendance

	Customs College Attendance vs.
	No Immediate College	Non-Selective 4-Yr College	Selective 4-Yr College	Highly Selective 4-Yr College
Female person	1.343 *** (0.081)	1.158 (0.116)	ane.317 ** (0.137)	i.154 (0.148)
Black	0.746 * (0.087)	0.903 (0.166)	0.368 *** (0.070)	0.422 *** (0.099)
Hispanic	0.437 *** (0.048)	0.662 * (0.114)	0.544 ** (0.111)	0.391 *** (0.082)
Other race	ane.205 (0.200)	0.812 (0.177)	two.158 * (0.728)	0.475 ** (0.116)
U.South. born	1.752 *** (0.169)	1.719 *** (0.266)	1.530 * (0.292)	1.406 + (0.246)
2d generation	one.887 *** (0.184)	1.451 * (0.236)	two.592 *** (0.462)	one.054 (0.209)
Intact family	0.974 (0.066)	1.023 (0.112)	one.173 (0.132)	1.185 (0.163)
Mother college graduate	i.071 (0.084)	one.011 (0.123)	0.925 (0.108)	0.865 (0.126)
Neighborhood social status	one.003 (0.047)	0.678 *** (0.052)	0.760 *** (0.063)	0.826 * (0.075)
Neighborhood non-poor	0.838 *** (0.041)	0.899 (0.070)	0.934 (0.075)	0.950 (0.095)
Neighborhood homeowner	1.015 * (0.007)	1.027 * (0.012)	0.993 (0.011)	1.028 + (0.016)
HS GPA	0.940 (0.050)	0.430 *** (0.038)	0.278 *** (0.026)	0.084 *** (0.011)
HS honors courses	0.894 ** (0.032)	0.896 * (0.043)	0.820 *** (0.039)	0.754 *** (0.037)
HS AP credits	0.675 *** (0.076)	one.047 (0.153)	0.936 (0.135)	0.799 + (0.102)
HS absences	0.988 ** (0.005)	1.008 (0.009)	1.025 ** (0.010)	1.015 (0.013)
HS special education	0.827 + (0.082)	1.021 (0.185)	3.192 *** (0.830)	3.769 ** (1.742)
Educ. resources	0.978 + (0.012)	0.964 + (0.019)	0.881 *** (0.018)	0.862 *** (0.022)
Parental communication	i.028 (0.024)	0.936 (0.038)	0.982 (0.041)	0.950 (0.048)
Parental interest	0.992 (0.016)	ane.022 (0.029)	1.071 * (0.031)	1.141 *** (0.041)
Higher aspirations	ane.288 ** (0.100)	0.675 ** (0.081)	0.685 ** (0.086)	0.707 * (0.110)
College expectations	one.143 (0.093)	1.112 (0.140)	0.741 * (0.094)	0.401 *** (0.062)
Higher expectations (parental)	1.841 *** (0.228)	1.183 (0.265)	0.847 (0.225)	1.077 (0.333)
HS college-going rate	2.813 * (i.212)	one.073 (0.793)	0.840 (0.633)	half dozen.688 * (6.146)
Constant	0.089 *** (0.027)	10.620 *** (5.567)	94.290 *** (53.303)	1830.452 *** (1284.507)
LR χ 2	297.8	296.four	936.4	2325.3
P > χ 2	0.000	0.000	0.000	0.000
N	6,476	2,512	two,775	3,086

In Table 3, we report propensity score matching results of the handling furnishings for the treated under each counterfactual scenario using several culling matching algorithms. We restrict all analyses to the region of common support (α=0.01). As expected, this restriction results in the loss of very few cases for the comparisons generally closely matched on observed characteristics, i.e. betwixt community college and no immediate college or not-selective four-year higher; we lose more than cases when we compare community college goers to selective and highly selective four-yr college goers. Nosotros exercise not discover notable differences between our iii matching methods.

Tabular array three

Matching Estimates of Community Higher Attendance on College Completion

	CC vs. No Immediate College	CC vs. Non-Selective 4yr	CC vs. Selective 4yr	CC vs. Very Selective 4yr
Unmatched Differences	0.012 *** (0.004)	-0.079 *** (0.010)	-0.146 *** (0.010)	-0.513 *** (0.013)
Nearest Neighbor Matching (yard=ane)	0.016 *** (0.005)	-0.050 ** (0.020)	-0.100 *** (0.027)	-0.409 *** (0.093)
Nearest Neighbour Matching (k=5)	0.013 ** (0.005)	-0.058 *** (0.017)	-0.092 *** (0.024)	-0.348 *** (0.075)
Kernel Matching	0.012 ** (0.004)	-0.053 *** (0.015)	-0.091 *** (0.022)	-0.307 *** (0.061)
North (on common support)	6,471	2,512	2,675	3,008
% cases lost	0.ane%	0.0%	iii.six%	two.five%

Compared to students who attend four-year schools, community college students are less likely to complete a bachelor'south caste. Matching estimates propose a level of available'due south degree completion roughly 5 pct points lower for community college goers relative to students starting at a non-selective four-yr higher, merely larger penalties (9 to 10 percentage points lower) relative to those attending a selective four-yr college. ¹⁵ We observe a substantial community higher penalty relative to attending a highly selective 4-yr college: we discover a 41 percentage signal departure using single nearest neighbour matching, a 35 percentage betoken difference using nearest neighbor matching with five controls, and a 31 percentage point difference using kernel matching. These estimates suggest that the students most penalized by attention a community college are those with more than advantaged social backgrounds and better bookish preparation. It appears these students would be particularly better served by attending a highly selective four-year school, as we would await given the loftier graduation rates characteristic of selective colleges.

While we find penalties associated with community college attendance compared to attending a 4-year college, we also observe that customs college goers are significantly more than likely to obtain a available'south degree relative to students who do not immediately pursue postsecondary schooling. Nearest neighbour and kernel matching estimates suggest a level of bachelor'southward caste completion that is 1.3 to ane.6 pct points higher—a modest point increase but a large increase in the odds given the depression levels of bachelor'southward caste completion among this population. Thus, customs college attendance yields both a large punishment relative to attendance at a four-year higher, particularly highly selective ane, higher attendance, in improver to a modest do good relative to no firsthand postsecondary schooling. ¹⁶

In analyses available upon request we construct Rosenbaum bounds to appraise the sensitivity of our results to different levels of unobserved biases (i.e., different assumed relationships between potentially unobserved variables and the treatment). Unobserved factors that double the odds of attending community college would return the positive effects of attention compared to not immediately pursuing postsecondary education non-pregnant. The negative effect of attending community college compared to a four-year higher is not sensitive to additional unobserved factors that would (at least) triple the odds of going to a community higher. This finding may exist unsurprising given the large size of these negative treatments effects; however, we must behave in heed that these comparisons are certainly the most susceptible to violations of the selection on observables assumption.

Auxiliary Analyses of Effect Heterogeneity

In Table four, we decompose the counterfactual condition by propensity score strata, in order to assess the likelihood of each of the alternative counterfactual conditions for the typical community higher goer. Nosotros generate these propensity score strata by obtaining predicted values based on a logistic regression model predicting community college omnipresence relative to no customs college attendance. Amidst individuals with a high propensity for community college attendance only who did non attend community college, the majority did not enroll in any college within ane year of high schoolhouse graduation. In other words, for the majority of community college goers, the alternative to community higher attendance is non to get to college; the treatment furnishings reported in the first cavalcade of Table 3 therefore correspond to the largest proportion of the community college population. Amidst students with a depression propensity for community higher omnipresence but who did not attend community college, we find proportionately higher levels of 4-year college goers. The selectivity of four-year degree attendance increases every bit the propensity for community college attendance decreases. Thus the treatment effects reported in the second through fourth columns of Table 3 correspond to a smaller population of customs college goers.

Tabular array 4

Decomposition of Community College Alternatives by Strata-Specific Propensity for Community Higher Attendance

(propensity score)	Stratum 1 (0-.05)	Stratum ii (.05-.075)	Stratum iii (.075-.1)	Stratum 4 (.1-.15)	Stratum 5 (.15-.2)	Stratum vi (.2-.25)	Stratum seven (.25-.six)
No Immediate Higher	0.246	0.270	0.388	0.569	0.642	0.719	0.813
Non-Selective four-Year	0.060	0.076	0.089	0.096	0.111	0.098	0.098
Selective 4-Yr	0.112	0.187	0.186	0.144	0.151	0.134	0.056
Highly Selective 4-Year	0.582	0.467	0.336	0.192	0.096	0.050	0.033
N	552	540	515	1,357	1,683	1,557	one,518

To examine furnishings beyond the propensity for community college omnipresence, we fit separate nonparametric regressions of bachelor's degree completion on the propensity score for the treated and control groups and take the difference in the curves. The x-axis plots the continuous propensity score and the y-axis the differences in nonparametric regressions between treated and controls—i.due east., the treatment effect using the "smoothing-differencing" heterogeneous treatment effects method described in Xie, Make, and Jann (2013). We find a curvature in the trends in effects, with a negative effect for low-propensity individuals, leveling off to no upshot in the middle of the propensity distribution. ¹⁷ This result further demonstrates that the big negative effect of customs college attendance is only relevant for the relatively small population of community college goers with a low propensity for attendance.

Discussion and Conclusions

The interpretation of community colleges' role in stratification processes depends on the accurate assessment of the colleges' effects on educational attainment. Nosotros have shown that a thorough agreement of community college effects requires a clear specification of the probable alternatives to attending customs college for various subpopulations. With rich survey and administrative data from Chicago Public Schools, we use propensity score matching to study community higher effects. While some scholars accept expressed the potential for heterogeneous treatment furnishings, nosotros rigorously test for their presence. Nosotros find that attending to the complexity of the counterfactual status and how alternatives stand for to the propensity for community college yields a more accurate portrait as to who is penalized and who benefits from attending community college. The penalty to community college omnipresence is largest among students who would have attended selective four-year schools—students with advantaged social backgrounds and stiff academic preparation, who have a low propensity for community higher attendance. By dissimilarity, community higher omnipresence increases the likelihood of bachelor'southward caste completion among students who otherwise would not have attended college at all—students with disadvantaged social backgrounds and poor academic training, who take a high propensity for customs college attendance and represent the majority of the community college population. Without attending to such heterogeneity, researchers sweep aside violations to the ignorability supposition that influence the results, especially for particular subpopulations, and in this case enlarge the negative aggregate effect of customs college attendance. Indeed, the widespread notion of a negative community higher effect is conceivably driven past the unacknowledged violation of the selection on observables assumption inherent in past results comparing community college goers to attendees of selective iv-yr colleges.

A few caveats are in order. Starting time, we focus on a single urban context. Although doing and then potentially increases the internal validity of our results, our findings may yet not generalize to other areas of the United states. The Chicago Public Schools and City Colleges serve especially disadvantaged populations and both are widely known for their challenges with regard to funding and leadership. Second, the accuracy of our estimates hinges on whether we have captured all relevant observables that predict community college omnipresence and available's degree completion. The assessment of variation in effects by the counterfactual condition could by biased equally a result of unobserved selection if such option differs systematically across the distribution of groups and influences caste completion. As we note higher up, the bias is ostensibly largest when attendance is an unlikely event—i.e., for the comparing betwixt community college goers and those who attend highly selective schools, in which nosotros find the largest penalty to community college attendance. Indeed, the disaggregation of alternative counterfactual paths and its relation to the estimated propensity for community college attendance highlights the contrasts for which we may look the largest selection biases to operate. 3rd, we examine only one outcome: available's degree completion. Customs colleges serve many functions for a diverse population of students. Even those students who we discuss as existence penalized because they did not complete a iv-twelvemonth degree may nevertheless have benefited from community college attendance in the labor market, in the marriage market, in their social-psychological wellbeing and cocky-acceptance, in their social interest, and in other outcomes that signal life chances (Hout 2012; Rose 2012). Future research should go on to explore the broader affect of community higher attendance while attending to heterogeneity in effects.

Having made these cautionary statements, we believe our findings have of import implications for policymakers, practitioners, and researchers in higher education. Assuming a homogenous community college effect masks variation in effects across the distribution of college goers. Our analyses suggest that accurately describing the part that community colleges play in social stratification requires analyzing effect heterogeneity and the processes through which heterogeneity arises. We find a punishment to community higher omnipresence for advantaged students who have a low propensity for community higher and might take instead attended a selective college. Thus while it may exist truthful that some students would be ameliorate served past attending four-yr rather than two-yr colleges, our analyses suggest that relatively few students would accept washed so. Nevertheless, in the electric current era of widespread economic distress facing families alongside ascent college tuition costs, we may detect an increasing number of students who attend customs rather than four-year colleges, even highly selective ones. We cannot predict, however, that this trend volition entail a greater number of students facing a punishment from attention community college, given considerable concern over option bias for this population comparing, and since a shift in the population composition of community college students could coincide with a corresponding shift in effects. Conversely, we discover a small-scale benefit to community college attendance amongst disadvantaged students who accept a high propensity for community college and for whom attendance at a 4-year school was improbable. Indeed, the most likely culling to community college attendance is no immediate college attendance. Discussions amid academics and policy analysts should move beyond broad characterizations of the community college as a site of lost opportunities to addressing the means in which we can ensure that these schools are equipped to serve the large numbers of students for whom they are the chief and best option.

​

Heterogeneous Community College Effects on Bachelor'south Degree Completion

Acknowledgments

Nosotros thank Stephen Morgan, Ruth Lopez Turley, Dan Koretz, Jenny Nagaoka, Melissa Roderick, and several anonymous reviewers for helpful comments and suggestions. Versions of this article were presented at the American Sociological Association (ASA) 2012 Annual Meeting and the Social Demography Inquiry Group at UCLA. The first author benefited from facilities and resources provided by the California Center for Population Research at UCLA (CCPR), which receives cadre support (R24-HD041022) from the Eunice Kennedy Shriver National Establish of Kid Health and Human Development (NICHD). Support for the second and third authors was provided by an NAEd/Spencer Foundation postdoctoral fellowship to the tertiary author. We thank the Consortium for Chicago School Inquiry (CCSR) for providing access to the data. The ideas expressed herein are those of the authors.

Biography

•

Jennie E. Brand is Associate Professor of Folklore at the Academy of California–Los Angeles and Associate Managing director of the California Center for Population Enquiry. Her inquiry centers on inequality and its implications for various outcomes that bespeak life chances, and on causal inference and the application and innovation of statistical models for panel data. Electric current inquiry projects include evaluation of heterogeneity in the effects of didactics on socioeconomic outcomes and the social-psychological and socioeconomic consequences of disruptive events, such as job displacement.

Fabian T. Pfeffer is a Inquiry Assistant Professor at the Survey Enquiry Center and a Kinesthesia Affiliate at the Population Studies Center of the Institute for Social Research, Academy of Michigan. His enquiry focuses on educational inequality and social mobility and their institutional contexts. Ongoing projects include the comparative report of intergenerational wealth effects, multigenerational social mobility, and the comparative cess of the role of education in mobility processes.

Sara Goldrick-Rab is Professor of Educational Policy Studies and Sociology at the University of Wisconsin–Madison. She is the Founding Director of the Wisconsin Promise Lab, a translational research laboratory focused on improving college access for all students. Recent papers have articulated agendas for the reform of community colleges, documented causal impacts of financial aid on college outcomes, and explored the role of familial poverty in college choices.

Table A.one

Logistic Regression Estimates of Community College Attendance on Higher Completion

	Main guess	Excluding selective colleges
Treatment	0.37 *** (0.058)	one.186 (0.209)
All other controls	incl.	incl.
N	nine,533	7,216

Footnotes

¹Other studies have gone further in an effort to identify "true" comparison groups. For example, Melguizo, Kienzl, and Alfonso (2011) compare the college outcomes of community college transfer students and rising juniors from four-year colleges. It is perhaps unsurprising that they notice no negative community college effect, since theirs is, rather than a community college upshot, the issue of being a community college transfer educatee—a selective and high achieving grouping of customs higher attendees.

²Distance to higher is an invalid instrument if it affects four-yr degree completion directly, rather than simply indirectly through community higher omnipresence. Indeed, proximity to iv-year colleges increases the likelihood of attention 4-yr colleges, as it is frequently used equally an instrument in studies of the effects of iv-year college attendance (eastward.g., Carneiro, Heckman, and Vytlacil 2011).

³The subpopulation induced into treatment, which cannot actually exist identified, can differ on both observed and unobserved characteristics from the treated and the untreated populations. The TT is a combination of the upshot for individuals induced by the instrument (then-called "compliers") and individuals who are treated regardless of the inducement ("always-takers"); likewise, the TUT is a combination of the effect for compliers and individuals untreated regardless of the inducement ("never-takers"). Since an 4 is non directly informative about effects on always-takers and never-takers, instruments practise not normally capture the average causal effect on all of the treated or on all of the not-treated (Angrist and Piscke 2009). Moreover, those induced into treatment tin can differ as the inducement changes, because different instruments will bear on handling condition for different segments of the population (Angrist and Pischke 2009; Brand and Simon Thomas 2013; Gangl 2010).

^fourCollege selectivity is defined by Barron'south Profiles of American Colleges (2003), which categorizes colleges according to Sat scores, grade betoken average, grade rank required for admission, and overall admissions acceptance charge per unit. Colleges in the top two categories of Barron's Profiles, "Most Competitive" and "Highly Competitive," are considered highly selective for our purposes.

⁵Futurity enquiry that aims to more fully nourish to the complexity of pupil postsecondary pathways may still need to engage in empirical reduction (e.grand., through latent class or sequence assay) to identify a manageable and meaningful set of common omnipresence patterns. This is beyond the scope of the current report.

⁶Prior research using a regression framework estimates ATEs only if the assumption of effect homogeneity is true. That is, from our reading, no subpopulation weights are practical. If there is effect heterogeneity, the handling upshot in such models is neither a TT nor an ATE, simply a detail weighted average lacking a direct analog to the ATE or TT (or TUT) (Elwert and Winship 2010). Past inquiry using an IV framework yields local boilerplate handling effects (LATE), which are too not direct comparable to either the TT or ATE.

⁷City Colleges of Chicago is a network of 7 institutions with the aforementioned tuition and fees, serving a clientele that is more than over 70 percentage racial and indigenous minorities. Get-go-year freshman retention rates range from forty to lx percent (world wide web.college-insight.org). Co-ordinate to ipeds data from 2009-2010, enrollment ranges from roughly 5,000 to 13,000 students, with five of the 7 colleges having 7,000 to ix,000 students.

^viiiThe most pop customs colleges for CPS students are the Wilbur Wright, Richard J. Daley, and Harold Washington Metropolis Colleges. The about popular four-yr colleges are the University of Illinois at Chicago (highly selective), Northeastern Illinois University (not-selective), the University of Illinois at Urbana-Champaign (highly selective), Chicago State Academy (selective), Northern Illinois Academy (selective), Columbia College of Chicago (non-selective), and Southern Illinois University at Carbondale (selective).

⁹Neighborhood social status is the standardized hateful of the pct of persons sixteen years onetime or older who are managers and executives in a Census block and the (logged) mean level of education amid people older than 18 years. Neighborhood unemployment and poverty is based on the percent of males over 18 who were employed one or more than weeks during the year and the percent of families above the poverty line in a Census block. Neighborhood homeownership is the average number of years of tenancy of homeowners in the census cake.

¹⁰The National Center for Educational activity Statistics apply twelve months as the established time to brainstorm postsecondary schooling without delay. Nationally, median time to available's degree is about four years for those pursuing postsecondary schooling without filibuster, and seven years for those delaying beyond twelve months.

¹¹Given the low share of these combined cases (2.2 percent of CPS graduates for 2001) and the absence of potent arguments why this particular kind of omission should exist related to our outcome of involvement, we consider it unlikely that this exclusion biases our estimates.

¹²We accept complete data for basic demographic characteristics and a negligible share of missing information for a scattering of variables, such as Census Tract data (0.3 per centum), high school's college going charge per unit (0.7 percent), and students' high school GPA, number of honors classes, and AP credits (2.0 pct). Only our survey measures have a meaning share of missing information, but the reliability of these imputations is maximized by using 9^thursday form survey information to impute for missing eleven^th form survey information (i.e., we retain students who did not respond to the eleven^thursday grade survey but did provide information on the same questions in the 9^th grade; N=3,977). Those not responding to any survey are more than likely to be male person, Hispanic, and from a less reward neighborhood.

¹³While we do not eliminate from the sample those who take stated they do not intend to obtain a bachelor's caste, we do compare indivduals with like expectations and aspirations during secondary schoolhouse. While educational expectations are malleable and should not define the sample, they may signal similar personality characteristics or ambitions that predict college pathways and could bias observed relationships between community college attendance and bachelor's caste completion. However, as these variables are potentially endogenous, nosotros have likewise estimated our main analyses without workout on college aspirations and expectations as a sensitivity test and find that workout has no substantive impact upon our results. These results are available from the authors upon request.

^fourteenA naïve interpretation of a customs college effect (i.e., without attention to the range of alternative educational choices) that is based on a logistic regression of available's degree completion on community higher attendance and the full set up of controls suggests that attending community college lowers the odds of completing a bachelor'southward degree past 63 percentage. These results are reported in the kickoff column of Table A.1 in the appendix.

¹⁵If we restrict attention to those students who obtain a available's degree, over one-half of customs college students graduate from highly selective colleges while only about xiii percent of students who begin at not-selective and selective four-yr colleges graduate from highly selective colleges.

¹⁶If nosotros drop all those students who attend selective colleges from the simple regression models reported in the first column of Tabular array A.1 in the appendix, we eliminate the negative effect of community college omnipresence on available's degree completion. We study these results in the 2nd cavalcade of Table A.1.

¹⁷We likewise guess furnishings within strata (level-1) so the trends in furnishings (level-2) (using the stratification-multilevel" heterogeneous treatment furnishings method described in Xie, Brand, and Jann (2013)). We find a pregnant positive level-2 slope, indicating that the outcome of community college increases (the negative effect decreases) as the propensity for community college attendance increases. The result of attending community college compared to not attending community college on available'south degree completion is significant in the depression propensity score strata, with equally much as a 44 percentage subtract in bachelor's degree completion in stratum 1, but insignificant in the high strata. We also test a quadratic term for level-two, and find a meaning curvature to the tendency in effects. That is, we find that the negative effect of customs college attendance on four-year degree completion decreases (becomes less negative) every bit the propensity for community college increases, and and so flattens to no event in the middle of propensity score distribution.

Contributor Information

Jennie E. Brand, University of California–Los Angeles.

Fabian T. Pfeffer, University of Michigan.

Sara Goldrick-Rab, University of Wisconsin–Madison.

References

• Alba Richard D, Lavin David Due east. Community Colleges and Tracking in Higher Education. Sociology of Didactics. 1981;54:223–37. [Google Scholar]
• Alfonso Mariana. The Impact of Community College Attendance on Baccalaureate Attainment. Research in College Education. 2006;47(eight):873–903. [Google Scholar]
• Alon Sigal, Tienda Marta. Assessing the 'Mismatch' Hypothesis: Differences in College Graduation Rates by College Selectivity. Sociology of Education. 2005;78(iv):294–315. [Google Scholar]
• American Association of Community Colleges. Customs College Fast Facts. 2011 http://www.aacc.nche.edu/AboutCC/Documents/FactSheet2011.pdf.
• Bailey Thomas, Belfield Clive. The Benefits of Attending Community College: A Review of the Evidence. Community Higher Review. 2011;39(1):46–68. [Google Scholar]
• Make Jennie Eastward. Civic Returns to Higher Education: A Note on Heterogeneous Effects. Social Forces. 2010;89(2):417–434. [PMC complimentary article] [PubMed] [Google Scholar]
• Brand Jennie E, Halaby Charles N. Regression and Matching Estimates of the Effects of Elite College Attendance on Educational and Career Achievement. Social Scientific discipline Research. 2006;35:749–770. [Google Scholar]
• Brand Jennie E, Simon Thomas Juli. Causal Effect Heterogeneity. In: Morgan Stephen L., editor. Handbook of Causal Assay for Social Inquiry. New York: Springer; 2013. pp. 189–214. [Google Scholar]
• Brand Jennie E, Xie Yu. Who Benefits Nearly From College? Evidence for Negative Pick in Heterogeneous Economic Returns to Higher Instruction. American Sociological Review. 2010;75(ii):273–302. [PMC free commodity] [PubMed] [Google Scholar]
• Brint Steven, Karabel Jerome. The Diverted Dream: Community Colleges and the Promise of Educational Opportunity in America, 1900–1985. Oxford: Oxford University Press; 1989. [Google Scholar]
• Carneiro Pedro, Heckman James J, Vytlacil Edward. Estimating Marginal Returns to Education. American Economic Review. 2011;101(6):2,754–2,781. [PMC free article] [PubMed] [Google Scholar]
• Century Foundation Report. Bridging the Higher Education Split up: Strengthening Community Colleges and Restoring the American Dream. New York: The Century Foundation Printing; 2013. [Google Scholar]
• Clark Burton A. The 'Cooling-out' Role in Higher Pedagogy. American Journal of Folklore. 1960;65(half dozen):569–576. [Google Scholar]
• Cohen Arthur Thou, Brawer Florence B. The American Community College. San Francisco: Jossey-Bass Publishers; 1982. [Google Scholar]
• Dale Stacey, Krueger Alan B. Estimating the Return to College Selectivity over the Career Using Authoritative Earnings Data. National Bureau of Economic Inquiry; 2011. Working Paper No. 17159. [Google Scholar]
• Dougherty Kevin J. The Contradictory College: The Conflicting Origins, Impacts, and Futures of the Community College. Albany, NY: Land University of New York Press; 1994. [Google Scholar]
• Doyle William R. The Outcome of Community College Enrollment on Bachelor's Degree Completion. Economics of Educational activity Review. 2009;28:199–206. [Google Scholar]
• Goldrick-Rab Sara. Challenges and Opportunities for Improving Customs College Student Success. Review of Educational Research. 2010;80(3):437–469. [Google Scholar]
• Goldrick-Rab Sara, Pfeffer Fabian. Across Access: Explaining Social Grade Differences in College Student Mobility. Sociology of Education. 2009;82(2):101–125. [Google Scholar]
• Grubb Due west Norton. The Decline of Community College Transfer Rates: Evidence from National Longitudinal Surveys. Journal of Higher Education. 1991;62(ii):194–222. [Google Scholar]
• Guess Andy. A Penalty for Starting at a Community College? Inside Higher Ed. 2008 October 1; http://www.insidehighered.com/news/2008/10/01/pathway.
• Heckman James J, Ichimura Hidehiko, Todd Petra E. Matching every bit an Econometric Evaluation Calculator: Evidence from Evaluating a Job Preparation Programme. The Review of Economical Studies. 1997;64(4):605–654. [Google Scholar]
• Heckman James J, Urzua Sergio, Vytlacil Edward. Understanding Instrumental Variables in Models with Essential Heterogeneity. Review of Economic science and Statistics. 2006;88:389–432. [Google Scholar]
• Holland Paul Westward. Statistics and Causal Inference. Periodical of American Statistical Association. 1986;81:945–sixty. with discussion. [Google Scholar]
• Hout Michael. Rationing College Opportunity. The American Prospect. 2009 Oct 22; 2009. [Google Scholar]
• Hout Michael. Social and Economical Returns to College Education. Annual Review of Sociology. 2012;38:379–400. [Google Scholar]
• Kalogrides Demetra, Grodsky Eric. Something to Autumn Dorsum On: Community Colleges every bit a Prophylactic Cyberspace. Social Forces. 2011;89(3):853–878. [Google Scholar]
• Kane Thomas J, Elena Rouse Cecilia. Labor Market Returns to 2- and Iv-Year Colleges. American Economic Review. 1995;85(three):600–614. [Google Scholar]
• Karabel Jerome. Community Colleges and Social Stratification. Harvard Educational Review. 1972;42:251–62. [Google Scholar]
• Lechner M. Identification and Estimation of Causal Effects of Multiple Treatments under the Conditional Independence Assumption. In: Lechner M, Pfeiffer F, editors. Econometric Evaluation of Labour Market Policies. Heidelberg: Physica; 2001. pp. 1–18. [Google Scholar]
• Leigh Duane Due east, Gill Andrew M. Do Community Colleges Actually Divert Students from Earning Bachelor's Degrees? Economics of Education Review. 2003;22(one):23–30. [Google Scholar]
• Long Bridget T, Kurlaender Michal. Do Community Colleges Provide a Feasible Pathway to a Baccalaureate Caste? Educational Evaluation and Policy Analysis. 2009;31(1):30–53. [Google Scholar]
• Melguizo Tatiana, Kienzl Gregory, Mariana Alfonso. Comparing the Educational Attainment of Community College Transfer Students and Four-Year Rise Juniors Using Propensity Score Matching Methods. Periodical of College Education. 2011;82(3):265–291. [Google Scholar]
• Morgan Stephen 50. On the Edge of Delivery: Educational Attainment and Race in the United States. Palo Alto, CA: Stanford University Press; 2005. [Google Scholar]
• Morgan Stephen, Harding David. Matching Estimators of Causal Effects: Prospects and Pitfalls in Theory and Practice. Sociological Methods and Enquiry. 2006;35(1):3–60. [Google Scholar]
• Morgan Stephen, Winship Christopher. Counterfactuals and Causal Inference: Methods and Principles for Social Research. 2d. Cambridge, UK: Cambridge University Printing; 2014. [Google Scholar]
• National Heart for Education Statistics. Baccalaureate and Across Longitudinal Study. U South Department of Instruction. 2009 http://nces.ed.gov/datalab/tableslibrary.
• Reynolds C Lockwood. Where to Attend? Estimating the Effects of Kickoff Higher at a Two-Year Institution. Economics of Education Review. 2012;31(4):345–362. [Google Scholar]
• Reynolds Curtis Lockwood, DesJardins Stephen L. The Use of Matching Methods in Higher Education Research: Answering Whether Attendance at a Two-Twelvemonth Institution Results in Differences in Educational Attainment. Higher Education: Handbook of Theory and Research. 2009;24:47–97. [Google Scholar]
• Reynolds John, Stewart Michael, MacDonald Ryan, Sischo Lacey. Have Adolescents Become Besides Aggressive? High School Seniors' Educational Plans, 1976–2000. Social Issues. 2006;53(2):186–206. [Google Scholar]
• Roderick Melissa, Coca Vanessa, Nagaoka Jenny. Potholes on the Road to College: High Schoolhouse Effects in Shaping Urban Students' Participation in Higher Application, Four-Year College Enrollment, and College Match. Sociology of Instruction. 2011;84(iii):178–211. [Google Scholar]
• Rose Mike. Back to School: Second Chances at Higher Education. New York: The New Printing; 2012. [Google Scholar]
• Rouse Cecilia E. Democratization or Diversion—The Issue of Community Colleges on Educational Attainment. Journal of Business organisation and Economic Statistics. 1995;thirteen(2):217–224. [Google Scholar]
• Sandy Jonathon, Gonzalez Arturo, Hilmer Michael. Alternative Paths to College Completion: Outcome of Attending a 2-Year Schoolhouse on the Probability of Completing a Four-Year Degree. Economic science of Instruction Review. 2006;25(5):463–471. [Google Scholar]
• Shavit Yoshi, Arum Richard, Gamoran Adam., editors. Stratification in Higher Instruction: A Comparative Study. Palo Alto, CA: Stanford University Press; 2007. [Google Scholar]
• Shaw Kathleen, Goldrick-Rab Sara, Mazzeo Christopher, Jacobs Jerry. Putting Poor People to Piece of work: How the Work-First Idea Eroded Higher Admission for the Poor. New York: Russell Sage; 2006. [Google Scholar]
• Turley Ruth N López. Higher Proximity: Mapping Admission to Opportunity. Sociology of Educational activity. 2009;82(2):126–146. [Google Scholar]
• Whitaker David G, Pascarella Ernest T. 2-year College Attendance and Socioeconomic Attainment: Some Additional Evidence. Journal of Higher Instruction. 1994;65(two):194–210. [Google Scholar]
• Xie Yu, Brand Jennie Eastward, Jann Ben. Estimating Heterogeneous Handling Effects with Observational Data. Sociological Methodology. 2012;42:314–347. [PMC free commodity] [PubMed] [Google Scholar]

spencercultin.blogspot.com

Source: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4375965/

Share this post