Failing at Remediation? College Remedial Coursetaking, Failure and Long-Term Student Outcomes

Abstract

Colleges offer remedial coursework to help students enrolling in post-secondary education who are not adequately prepared to succeed in college-level courses. Despite the prevalence of remediation, previous research presents contradictory findings regarding its short- and long-term effects. This paper uses a doubly robust inverse probability weighting strategy to examine whether the degree completion and wage outcomes associated with remedial education vary by passing or failing remedial coursework. Using the NLSY Postsecondary Transcript-1997 data, we find that almost 30% of remedial course takers fail a remedial course. Students who took and passed their remedial coursework at both 2-year and 4-year colleges were more likely to graduate from college than similar students who did not take remediation. For both 2-year and 4-year college entrants, students who failed remedial coursework were less likely to obtain a bachelor’s degree and, among degree receivers, took longer to graduate. Students who entered 2-year or 4-year colleges and who failed remedial coursework earned lower wages over time compared to similar students who never took remediation. Among 4-year college entrants, these wage differences seem to be explained completely by degree completion. However, wage differences for 2-year college entrants still remain after accounting for degree receipt. Our findings thus suggest that while many students may benefit from remedial education, a substantial number of students struggle with remedial coursework and fail to realize the intended benefits.

Appendix

Doubly Robust Inverse Probability Weighting

In the first step of doubly robust IPW, we estimate treatment propensities (P) for each student. Using covariates discussed in the paper, a propensity score is estimated for each student. An individual variable does not have to be a statistically significant predictor of treatment to be included in the propensity model since the objective is for students in the treated and control categories to be balanced on the covariates. The propensity scores are estimated using a multivariate logistic regression model predicting the probability of a student receiving the treatment (i.e., not taking a remedial course, taking and passing a remedial course, and taking and failing a remedial course). All covariates discussed in the paper were included in the multiple logistic regression equation to predict the probability of treatment:

$$Pr\left( {Remedial\,Group} \right)_{i} = { }\alpha_{i} + \beta_{k} {\varvec{X}}_{ki} + \user2{ }\varepsilon_{i}$$

(2)

Equation (2) predicts the probability of student i being in one of three groups: never took a remedial course, took and passed remedial coursework, and took and failed a remedial course. ${{\varvec{X}}}_{ki}$ is a vector of control variables.

We estimate each student’s predicted probability of being in each of the remedial groups in Eq. (2), and use these probabilities to create inverse probability weights. For each treatment category t (never took remediation, took and passed remediation, or took and failed remediation), we define our inverse probability weight as:

$$W = 1/\hat{P}_{t}$$

(3)

where ${\widehat{P}}_{t}$ is the predicted probability that a student received the treatment that he or she received.

For doubly robust IPW estimators, the same covariates used to estimate the probability weights for Eq. (2) are also included as controls in a linear probability model predicting our degree and wage outcomes. We estimate two sets of these models; the first set predicts bachelor’s degree completion in any field and the second set predicts the average wage outcomes. Thus, our first model predicts whether students complete a bachelor’s degree as a function of being in one of the three remedial groups: never took remediation, took and passed remediation, and took and failed remediation:

$$\Pr \left( {Bachelors} \right)_{i} = { }\alpha_{i} + { }\beta_{1} Remedial_{i} + { }\beta_{k} {\varvec{X}}_{ki} + \user2{ }\varepsilon_{i}$$

(4)

where Remedial_i is a dummy variable equal to one if a student ever took remediation and zero otherwise and X_i is a vector of background controls for doubly robust estimates. To estimate the relationship between failing remedial coursework and our other outcomes, we estimate additional models that take the same general form as (4), but instead of Remedial_i, we use a dummy variable equal to one if a student took and passed their remedial coursework and zero otherwise, or alternatively a dummy variable equal to one if a student took and failed their remedial coursework and zero otherwise. The error term, ${\varepsilon }_{i}$ captures characteristics not accounted in the model that influence the outcome variable. We estimate these models separately for students who entered a 2-year and 4-year college and we use similar models to predict average post-college wages for the latest 5 years (2007 through 2011) (Tables

Table 6 Linear probability models (LPM) predicting failure among remedial course takers

6,

Table 7 Doubly robust estimates of outcomes associated with remedial course failure relative to passing remedial coursework

7).