Read on Twitter
We Want You Back: Uncovering the Influences on In-Person Instructional Operations in Fall 2020 - This paper is ready to be viewed and follow this thread for some to the point details!
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3778772
You may know a few months ago my team generated a descriptive analysis to examine the connections between various political and institutional factors and reopening in-person for Fall 2020.
This was a GREAT first step but we could do better...
https://preprints.apsanet.org/engage/apsa/article-details/5f7f5d9eefc0c2001974ecc3
This was a GREAT first step but we could do better...
https://preprints.apsanet.org/engage/apsa/article-details/5f7f5d9eefc0c2001974ecc3
To expand, we decided to use Structural Equation Modeling to test the outcome of being In-Person by Sept 9. Didn't have access to decisions at the time when classes started for the individual institutions... as detailed in the paper we're confident this represents said decision.
We also expanded variables by including more state sociopolitical measures - introducing county sociopolitical measures AND calculating COVID (State+county) per capita at the time of last known decision and not in aggregate at our gatekept time (as we previously did).
The structure of the model - we were guided by work like Hopkins (2018) that suggests counties have lost sociopolitical independence from wider/stronger influences like states. We tested various directions between state and county features and this one was the strongest.
To be noted, the ovals are latent variables and measure the "unobserved" that predict our observed variables ( squares). So the ovals are not "exactly" what the squares are - it is a construct.
The variables that are "exactly" what is observed are the squares and rectangles.
The variables that are "exactly" what is observed are the squares and rectangles.
Back to the model, given policy preferences, sociopolitical features were pathed to State Revenue Declines and Pandemic Severity - providing indirect effects through to In-Person instruction. Same with Pandemic Severity through Revenue Declines (why we do SEM).
The model below is the same as above with the direct effects noted. The models are of strong fit (you can see them in the paper).
So lets talk some interesting trends in the underlying structure - being everything to the left of In-Person Instruction or in this table everything until In-Person Instruction...
Pay attention to State (r=.40) and County Sociopolitical Features on Pandemic Severity (r=-.23).
Pay attention to State (r=.40) and County Sociopolitical Features on Pandemic Severity (r=-.23).
Despite being predicted by similar sociopolitical features there are divergent trends. At first we thought, "counties that house institutions in our sample may be more politically independent from state 'in-group' pressures."
That'd be great - independence at the local level!
That'd be great - independence at the local level!
We then thought, "What if those features encouraged institutions to make decisions earlier when COVID was lower?" Using prior research, we decided to look for correlations between the observed features of the county-level latent variable and a decision by June 1 which consists of
n=527 institutions (21%) of the sample of which n=484 (91% of early deciders) were In-Person.
We found +correlations between the observed county-level factors and making this early decision; therefore, suggesting alignment of larger GOP policy preferences and not independence.
We found +correlations between the observed county-level factors and making this early decision; therefore, suggesting alignment of larger GOP policy preferences and not independence.
Ok, for our overall model which was developed using all institutions in our sample - County Features (r=.13), Pandemic Severity (r=-.10), and State Features (r=.09) were significant influences on in-person instruction. Hard to tell which is really stronger...
As institutional sectors are expected to respond uniquely, we test subgroups in the larger model (what you do in SEM)
Pandemic Severity mattered overall, it was not significant for 4-year or 2-year public institutions (red).
Yet 4-year privates were sensitive to it (green).
Pandemic Severity mattered overall, it was not significant for 4-year or 2-year public institutions (red).
Yet 4-year privates were sensitive to it (green).
While 4-year privates should be applauded, state (yellow) and county (blue) features were more impactful - especially county features. There is not enough research out there showing how 4-year private institutions are politically sensitive and that is one major contribution here.
We believe outcomes for 4-year private institutions illustrate a desire to remain close to the sociopolitical "in-group" to maintain political favorability and not alienate the enrollment pool - or to lose more students to local competitors (who also desire in-group proximity).
The other outcomes are a bit less, exciting but important.
4-year publics are dependent on state power so they're more sensitive to the state and 2-year publics dependent on both state and county power, but are more sensitive to local sociopolitical features.
4-year publics are dependent on state power so they're more sensitive to the state and 2-year publics dependent on both state and county power, but are more sensitive to local sociopolitical features.
So, what does this all mean? Sociopolitical preferences > COVID when reopening + ain't a good look for public institutions in the sub-sector analyses.
4-year privates are more sensitive to sociopolitical features that I think most of us may know or have previously captured.
4-year privates are more sensitive to sociopolitical features that I think most of us may know or have previously captured.
But Dan... bro... what about the institutional features they MUST matter. Yea, we figured that too and tested this model.
It produced some BAD fits so it was scrapped.
It produced some BAD fits so it was scrapped.
We tested another model that has the "Prestige Elements" latent variable out of the model and it produced acceptable fits but statistically weaker fits than our reported model.
We believe two things...
We believe two things...
1. This other model could be better for Spring 2021 tests given a variety of factors stated in the paper.
2. Maybe these factors are better tested as subgroups within the overall model (something, I'm doing now).
2. Maybe these factors are better tested as subgroups within the overall model (something, I'm doing now).
To wrap up, generally, Sociopolitical Preferences > COVID.
CC: The Team who make the dream
@FitzEdPolicy @madisonmdell @sam_snideman @ChrisMarsicano @rkelchen
+ who may be interested
@DTandberg @aefpweb @ASHE_CPPHE @imbernomics @cant_b @econroy_1 @dhfeld1 @Raeda_Anderson @TX2MS @rtoutkou @mikekofoed @mmilleradams @MichelleDimino
@FitzEdPolicy @madisonmdell @sam_snideman @ChrisMarsicano @rkelchen
+ who may be interested
@DTandberg @aefpweb @ASHE_CPPHE @imbernomics @cant_b @econroy_1 @dhfeld1 @Raeda_Anderson @TX2MS @rtoutkou @mikekofoed @mmilleradams @MichelleDimino
Also my #WWE gif game was ON POINT here...
Finally a special shout out to @TX2MS and @Raeda_Anderson for an excellent conversation on methods and making me think through a few more things!
