IRPE Colloquium: Predicting Graduation for First Time Full-time Freshmen at CSUCI

IRPE Colloquium: Predicting Graduation for First Time Full-time Freshmen at CSUCI

Kristin Jordan (SOC, IRPE) & Jared Barton (ECON)

4/23/18

Where to find info

See this IRPE research brief

Characteristics at admission predict graduation: HS GPA, SAT scores, high school curriculum, race/ethnicity, income, parent education. Looking at underrepresented (URM) student achievement gap, income gap (Pell grant or not), first generation college student or not. 

Use info at admissions to explain past graduation rates and also to forecast and understand future graduation rates. Goals: examine which characteristics predict student success, and to decompose achievement gaps into what we can explain and what is left unexplained in understanding those gaps.

Achievement gap characteristics overstated because students appear in more than one category. Achievement gap characteristics are correlated with other known (positive) predictors of graduation.  

Our students: their SAT and ACT-converted-to-SAT, are negatively correlated with URM status, but scores are positive predictor of performance and graduation. (2016 rolled out a new algorithm for the SAT, so need to account for that.) If have high enough GPA (over 3.0), don't need to submit SAT scores - this would create fewer high GPA students, so ordered by GPA to correct for it, turns out higher GPA take SAT and report it. 

Logistic regression for graduation rates in 4, 5, 6 years for full time first time freshmen as a function of gap characteristics, preparedness measures, and math/English proficiency, plus control variables. Looked at 8 models - for the first, looked at gap characteristics on their own. Model 2, looked at all together, models 3-5, add in preparedness measures. Models 6-8 account for control variables (undeclared major is correlated with less success; sex, veteran status, county of origin, etc.). 

4 year URM Achievement gaps URM v non = 7.5% less likely to graduate. (Only 22% currently graduating, so non-URM closer to 30% graduation rate). When use URM and Pell-eligible receiving, controlling for gap over lap controls for about 33%, down to 5%. Models 3, 4, and 5, controlling for SAT/ACT score and high school GPA; if compare to math not prepared.

Math alone has larger marginal effect than both Math and English - halves the gap to under 3%.

Pell eligible, URM, or first gen also positively correlated with being female, who are already more likely to graduate than dudes. Gap is artificially small because picks up women, then gap increases again. Our gap is 50% smaller than what we're actually reporting to Chancellor's office. 

Pell gap -9.33% (includes transfers - Pell gap for transfers is actually +1 - more likely to graduate than non-Pell eligible). Gap is almost halved once account for preparedness measures. It's the money gap doing most of the work though we spend most of our time on the URM gap. We don't have a first gen gap, we have a Pell gap: it's the low income, not the gen causing the gap. 

Me: Chancellor's office dictates the equation we use?

Barton: Yes. Wish it were not.

6 year gap (fewer data by definition, data drops between 33%, so standard error increases because sample size fell). URM achievemnt gaps fall when start controlling - little gap for URM students once account for preparedness (but not much gap in 6 year to begin with). For Pell, continues to be a sizable effect. First gen gaps (non first gen students catch up in years 5 and 6 in a way that first gen don't). 

Raw graduation rates through Spring 2017. Take coefficients from model, apply to all of the data including cohorts that can't have graduated yet. Forecasting: moving forward, flatlines at 25%. Keep doing what you've been doing, keep getting what you've gotten. Take gap variables and reduce size in calculation (20% smaller than previous year - coefficient on URM, Pell eligible receiving, and first gen), graduation rate rises. Bad news is goal is 40%. For 6 year story is better, we only have 4.5 percentage points to go - almost there. 4 year gap is a whole std deviation to close, way different than the much smaller 6 year gap.

Even with a magic bullet educational intervention...can't close the gap entirely and the 4 year gap is particularly difficult.

SO what are the factors we are not accounting for?

Ivona: Pell folks most likely to work a lot. Maybe easing that burden through online classes. Also based on old remediation model. Research says it's better to let people to swim in the pool than do the baby steps. Barton feels that is wrong. It's an imperfect test, the new model. 

Ivona: works elsewhere but totally different sample and preparedness. Apples to oranges. 

Barton: More comparable is CA community colleges.

Gap characteristics correlated positively w each other, negatively with positive predictors of graduation. Pell gap is largest and most consistent. First gen gap is smallest and least persistent. 

CI will get to roughly half of the 2025 graduation rate goals for FTFT if closes 100% of the unexplained URM, Pell, and first-gen gaps. In the 6 year, the URM gap is 1.6%, but never hits 0. In 4 year, closer to 3.5%. We should optimize organization for and focus on the 6 year goal - achievable and reasonable, especially since we should admit we admit students underprepared for college. We need to claim that, not to do so is stupid. 

Next analysis will be of transfer students. Transfer data is harder. Only one way to come in as FTFT, but way more interesting ways to come in as a transfer. All these interventions on campus: find them, find participants involved and some not, and look for treatment effect. Now design treatment and controls: here are the people we will intervene with. We need to actually assess intervention programs' success in closing the gaps. Further study to explain gaps (do you work, how many hours do you plan to work this week, get to food insecurity indirectly).

Grad rate same/persistent over time despite underpreparedness increasing - either we are doing more yeoman's work in teaching more and more underprepared students, or we are taking a fixed cut of the pie each year (declining population - same marginal effect, we're growing but overall decline). These data don't look at that.