AEFP 45th Annual Conference

Toward a Meaningful Impact through Research, Policy & Practice

March 19-21, 2020

Estimates of the Impact of Natural Disasters on Postsecondary Enrollment and Retention in the United States

Presenter: 
Alexa Cecil, West Virginia University, aantill@mix.wvu.edu

Communities across the United States experience a range of natural disasters, including Hurricane Katrina, a devastating tornado in Alabama, and recent wildfires in California. These incidents can be disruptive for college students (Brown, 2011; Whitford, 2018) and for colleges and universities that have responsibilities for disaster preparation and response (Tkachuk et al., 2018). With respect to planning and forecasting, these institutions need to be able to anticipate costs of future disasters accurately, and national or regional agencies that calculate disaster costs should be able to account for disruption in students’ educational trajectories that may influence future productivity.
Relevant literature includes evidence of natural disasters’ effects on education- and non-education-related outcomes in the U.S. and abroad. Researchers have found that nearby natural disasters have effects on students’ test scores and likelihood of college enrollment and the demographic profile of institutions’ enrolled students (Sacerdote, 2012; Cerqua & DiPietro, 2017; Spencer et al., 2016). Boustan et al. (2019) study natural disasters in the U.S. from 1930 through 2010 and find that severe disasters (causing over 25 deaths) increased county-level out-migration and reduced housing prices, but they find little effect from less-severe disasters. Boustan et al.’s (2019) county-level data are measured by decade, and the present study makes contributions by introducing annual disaster incidence and outcome measures and focusing specifically on outcomes relevant to U.S. colleges and universities.
The primary research question for this study will investigate the impact, if any, of a natural disaster in a college’s or university’s county in the preceding year on that institution’s total enrollment, first-time enrollment, and first-year retention. Additional research questions will consider any differences in impact based on disaster severity, disaster physical proximity or recency in time, and college or university characteristics, including control, degree level, spending on student services, and share of enrollment from out-of-state.
Data for this investigation come from the Federal Emergency Management Administration’s (FEMA) Disaster Declarations Summary, the Centre for Research on the Epidemiology of Disasters’ EM-DAT Emergency Events Database, and the National Center for Education Statistics’ Integrated Postsecondary Education Data System (IPEDS). The FEMA data include records of all emergency declarations in the United States by date, type, and county, and the EM-DAT data include the number of deaths from individual events and can be merged with the FEMA data. The merged disaster data can then be linked with the IPEDS data by county and academic year, allowing for identification of colleges and universities that experienced natural disasters within their counties in a given year and introduction of institution-level enrollment, retention, and spending measures. The analysis sample includes years from 2005 through 2016 and includes 1,236 disasters in 489 unique counties.
We estimate a model of the form
y_it=〖αX〗_it+δ_i+τ_t+ε_(it ),
where y_it is the outcome for institution i in year t, X_it is a vector of measures of natural disasters in proximity to institution i around time t, α are regression coefficients to be estimated, δ_i are institution fixed effects for each institution i, τ_t are year fixed effects for each year t, and ε_it is an error term. Across specifications, X_it can include indicators for any natural disaster in the previous year within institution i’s county, a natural disaster in institution i’s county in the previous year that caused at least 25 deaths, or an indicator outside of institution i’s county but within the same state in the past year.
Preliminary results suggest that first-time enrollment falls 1-3 percentage points at public two- and four-year colleges and universities in the year after a disaster in the same county, and total enrollment falls 1-2 percentage points at public four-year institutions. Retention falls by 0.5 percentage points at public four-year institutions with mean retention in the sample at 73 percent.
Next steps for this analysis include introducing measures of institutions’ spending on student services and share of student enrollment from out of state in order to investigate disaster sensitivity by resource level and reliance on enrollment of local students. Additional next steps include using GIS software to identify natural disasters in adjacent counties and introducing time lags to investigate differences in disaster impact by physical proximity and recency in time.
Findings from this study will assist both individual colleges and universities and local, regional, and national planning agencies to understand the costs of natural disasters for these institutions and identify any potential for disruptive effects on students, as well as to characterize aggregate costs for different kinds of institutions in the same county or region affected by a single natural disaster.

Poster: 

Comments

This is certainly a timely and important topic. I like the overall strategy. Two reactions: First, you might distinguish between institutions of various types/selectivity. One would expect little/no effects of natural disasters on enrollments (1st year or otherwise) at highly selective institutions. This should especially be true at institutions that enroll students from areas not affected by a disaster. For example, the recent tornadoes in Nashville may have no effects on enrollments at Vanderbilt. But, they may have negative effects on a local community college. Second, you may be able to augment your data on natural disasters using the SHELDUS data https://sheldus.asu.edu/SHELDUS/

Thank you for the feedback! Agreed on the point re: selectivity. We are trying to think of a few different measures of selectivity (acceptance rate, incoming test scores, Barron's competitiveness index, some combination) and check sensitivity of findings across. Selectivity applies less to community colleges but we're trying to think of relevant institutional characteristics like rural vs. urban as a proxy for population density (how many people might be affected by the disaster). I had not seen the SHELDUS data before so thank you for the heads-up. Our dataset is 2005-2016 (which I realized is not on the poster), with a sort of arbitrary decision to start at 2005, and I see the SHELDUS dataset starts at 2010. We might be able to use that to sanity-check our disaster data for 2010 forward.

Scratch that, the SHELDUS data goes back much further in time. Definitely will investigate and appreciate the lead.

Interesting and timely analysis! To the extent possible, thinking about the mechanisms through which we might expect different types of disasters to affect different types of post-secondary enrollment would really deepen this study. I imagine that the effect of a disaster might very on the type (e.g., hurricane vs. tornado) both directly but also through the differential effects disasters might have on short-term outcomes - that is, certain types of disasters may/may not have differential effects on K-12 school closures, the postponement of testing dates, etc., and those short-term effects may be driving the enrollment effects (e.g., students don't have as much time to spend on applications, they aren't scoring as high on entrance exams, etc). OR it could be that disasters serve as a signal in the college search process - this is a dangerous location where I shouldn't go to school. Since this study focuses on institutions' exposure to disaster, we might expect the analysis to primarily answer the second question, though could also capture the first mechanism if the institution primarily draws on local students. Excited to see the project advance!

Thank you very much for your feedback! We'll be able to go into greater detail on the theoretical framework in our paper vs. the poster. You have given us a lot to think about, especially relative to K-12 school disruption or distraction from the college search process. Signaling college local safety seems a natural one, and I would add potential local labor market disruptions to the list. Also agree that it will be important to check on differences across disasters that might imply different things for these different channels.

We have some additional evidence on New Orleans college outcomes (Harris and Larsen, 2018) at the Education Research Alliance for New Orleans web site. I think it would be worth distinguishing more between the effects on total enrollment in, say, a state, versus changes in the location of enrollment. I think you're mainly capturing the latter. There are two different questions. COVID will be different because it's basically everywhere, as opposed to hurricanes, which. are concentrated in specific locations. But there will be variation in intensity (COVID cases as a share of the population). I could imagine people shifting to small college towns and/or closer to home.

This is an important point. Getting at the intensity of disasters (to the extent possible) would be a wonderful addition to this research.

Thank you very much for this feedback! We have some preliminary results with intensity of disasters based on deaths. This follows the Boustan et al. (2019) NBER paper using a cutoff of 25, but they used disasters over decades. We use annual data. We have some significant results, but lots of 0s for individual counties. More observations for the adjacent counties or adjacent states. We'll look at sensitivity if we bring the cutoff down from 25. The SHELDUS data Dave Marcotte mentioned above also has information on injuries so that might also be a helpful measure.

Thank you very much for your feedback. I hadn't seen your work in New Orleans and I've downloaded that paper for us to review. Agree with the point on states as a better proxy for "total" enrollment than institutions or counties. I'm not sure how much we should try to think of these questions (even institutions and counties as we have done here) as close enough to be in one paper or use one motivation. Implications for colleges and universities of students going somewhere else or attending closer to home are clear if they lose tuition revenue, but there's also the macro question. The macro question might also be well served with something like ACS data in addition to what we're looking at. You have definitely given us some food for thought.

Thanks for sharing this interesting work! I agree with Dave that, when I first read the poster, I wondered what the theory of action was to look at disasters in the same county as a college. Most theories of college selection have to take into account that students who select to attend certain types of institutions (like more selective universities) are also more willing to travel. It could be useful in future iterations of the work to outline the theory of action and why the research focuses on only two-year publics, four-year publics, and four-year private not-for-profits. I will also mention I was a little confused about the measure of enrollment. IPEDS includes first-time degree seeking students broken down into 3 groups: all, full-time, and part-time. I'm assuming the overall measure was used here, but it would be helpful to be more specific about the exact type of first-time students being measured for the outcome. Thanks again and if I can be helpful in the future, let me know!

Thank you very much for this feedback! First, on the enrollment measure point, this is great. Our results now use the total measure from IPEDS. We should use FTE rather than the total even when we report aggregate results, but it would also be useful to break out FT students vs. PT students in detail. (As I'm thinking about your point, it's plausible that some students might switch from FT to PT and that wouldn't show up if we just look at total enrollment.) With the FEMA disaster data, we have to rely on the county as the unit of observation for disaster incidence, and then look at institutions in those areas. I agree with your comments and comments above that it will be helpful to disaggregate across institutions that rely relatively heavily on local student enrollment and those that do not.

I like these RQs. I'll second (or third, I suppose) the comment that getting to the mechanism will be most essential here. If this is really a disruption of marketing for specific institutions, it may be less of a big-picture policy concern but will definitely be a big deal and useful information for affected institutions to know.

Thank you very much for the feedback! We'll definitely take away from this feedback a charge to try to differentiate the story for individual institutions (and across dimensions like selectivity and relative reliance on enrollment of local students) from the policy story or more macro-level story. I think we have some good directions to go to try to investigate this.

Add new comment