Our study design and our findings have limitations, which should inform how they are interpreted and used. We list some of these limitations below.
 

Comparison Group Composition

While we use sophisticated techniques to assemble high-quality comparison groups and while we have extensive person-level data at our disposal in this task, we recognize that we cannot locate individuals for our comparison groups who perfectly mimic individuals in our treatment groups, minus only the choice to pursue postsecondary education during the follow-up period. In this regard, our comparison group earnings estimates are subject to some level of bias.
 

Unobservable Earnings

We have no ability to observe certain types of earnings of students in our treatment groups or of individuals in our comparison groups, notably out-of-state earnings.

This data limitation creates a dilemma for us when faced with students in our treatment groups or of individuals in our comparison groups who have no reported earnings in a given quarter or stretch of time. In these circumstances, we need to set rules for when to treat zero reported earnings, on the one hand, as evidence that the individual or student was unemployed or, on the other hand, as evidence that the student or individual in question was in fact employed but in a manner (including, for example, out-of-state employment) that did not register in the Texas UI wage records.

Faced with this dilemma, we take the position generally that when students in our treatment cohorts or individuals in our comparison groups have zero reported earnings in a quarter, it is because they were unemployed.

We depart from this default policy in one major way, as explained here, which is to exclude from our treatment cohorts and our comparison groups anyone who has no reported in-state earnings in any quarter late in the relevant follow-up period and who, by inference, has a heightened likelihood of being employed outside of Texas with material unobserved earnings.

We recognize that this exclusion policy might bias our findings. For example, it might lead us to exclude students from our entering cohorts who have no reported earnings in-state, not because they are employed out-of-state, but because they are unemployed in-state and who should therefore be included in our analysis.
 

External Validity

We cannot be certain that our findings generalize to students other than the ones we study directly.

For example, we only study entering cohorts from 2008-09 through 2018-19. We do not know with certainty whether findings from these cohorts in this era generalize to more recent cohorts that we have not studied yet.

In another example, in entering cohorts in 2008-09 and in the several entry years thereafter, we study only entering students who recently graduated from high school, for reasons we explain here. We do not know with certainty how well our findings for these entering cohorts generalize to older students in the same cohorts who we excluded from our analysis.
 

Cost Estimates

As explained here, we are limited in various ways in our ability to observe the costs that students incur from their choice to enroll, and we elect not to include in our cost calculations estimates of students’ expenses associated with housing, meals, transportation or childcare. These data limitations and choices might lead to measurement error in our cost estimates.

Over the next several years, we hope to repeat and improve the study that we publish here, in Texas and in other states. In future versions of our study, we expect to expand our research questions and to improve our methodology in the following ways.
 

Findings for Completers versus Non-completers 

We expect to report separately on outcomes for completers and non-completers. In this analysis, we will examine the outcomes of various sub-groups of non-completers (e.g., non-completers who exit early, non-completers who exit late, and non-completers who eventually earn a degree elsewhere). We will also study outcomes for completers and non-completers separately by degree type, by program, by demographic group, and by institution type. We are motivated to undertake this analysis because little is known empirically about the economic outcomes of non-completers.
 

Longer Follow-up Periods

We expect to extend the follow-up periods in which we study cumulative net VAE, especially for associate’s degree seekers and certificate seekers. One important question that this analysis might allow us to answer is when annual value-added earnings for associate’s degree seekers and certificate seekers stabilize.
 

Further Analysis of Immature Cohorts 

We expect to analyze in more detail the outcomes of immature entering cohorts that have not completed a follow-up period post-entry. From this analysis, we might learn more about when a cohort’s early results can be used to reliably predict its terminal cumulative net VAE at the end of its follow-up period.
 

Confidence Scoring of Estimates 

We expect to do work that will allow us to characterize the level of confidence that we have in various classes of cumulative net VAE estimates. 

How confident we are in the validity and accuracy of any given cumulative net VAE estimate is a function of a wide range of factors, including for example the size of the treatment cohort in question and the quality of its comparison group in question. In this analysis, we will assign confidence scores to our findings for various classes of cohorts (by entry year, by degree type, by program, and by demographic group).
 

Characteristics of High-VAE and Low-VAE Institutions

We expect to analyze how cumulative net VAE outcomes vary across types of institutions. In this analysis, we expect to show outcomes for institutions categorized by size, geography, per-pupil spending, selectivity, program focus, and other characteristics.
 

Additional Programmatic and Demographic Cohorts 

We expect to study additional types of programmatic and demographic cohorts. For example, we might study demographic cohorts organized by students’ English achievement in high school, by students’ IEP (individualized education plan) status in high school, and by the level of education of students’ parents. We might also study programmatic cohorts that are organized by new clusters of CIP codes.
 

Expanded Data Sources and Entry Years

With each repetition of the study, we expect to expand the entry years that we study and, if data sources allow, to expand the person-level data that we use. We will continue to seek data that allows us to observe individuals’ out-of-state earnings and out-of-state postsecondary education.
 

Discounting Estimates

In this release of the study, we adjust our cumulative net VAE estimates for inflation (all results are reported in 2023 dollars), but we do not discount the cash flows involved in students’ cumulative net VAE outcomes to account for the time-value of when they occur. We expect to do this in the next iterations of our study.