Read on Twitter
Our pre-print brief is available for download:Another Lesson on Caution in IDR Analysis:
Using the 2019 Survey of Consumer Finances to Examine Income-Driven Repayment and Financial Outcomes
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3758574
In this brief, @FitzEdPolicy, @ChrisMarsicano and I used the Survey of Consumer Finances '19 dataset to update the models we generated with the '16 dataset.
See the original here: https://preprints.apsanet.org/engage/apsa/article-details/5e8b3bedcf138e0019f49641
See the original here: https://preprints.apsanet.org/engage/apsa/article-details/5e8b3bedcf138e0019f49641
Given the timing of data collection, the SCF '19 dataset is more likely to include more people in REPAYE than the '16 dataset - so we were curious how that changed WHO may be enrolled in IDR.
What we found was... again, how you specify the model matters.
Again, we ran two types of models - ones inspired by the Collier (2020) paper and one inspired by Looney & Yannelis (2019).
https://ir.library.louisville.edu/jsfa/vol49/iss2/3/
https://www.brookings.edu/wp-content/uploads/2018/02/es_20180216_looneylargebalances.pdf
Again, we ran two types of models - ones inspired by the Collier (2020) paper and one inspired by Looney & Yannelis (2019).
https://ir.library.louisville.edu/jsfa/vol49/iss2/3/
https://www.brookings.edu/wp-content/uploads/2018/02/es_20180216_looneylargebalances.pdf
Descriptively, there are interesting (but not surprising) differences.
SCF19 - 35% of borrowers in IDR
SCF16 - 27%
SCF19 - IDR avg. Loan Balance $62k
SCF16 - $43k
SCF19 - IDR enrolled BA only 31%
SCF16 - 26%
SCF19 - IDR enrolled Household Wages Avg $70k
SCF16 - $62k
SCF19 - 35% of borrowers in IDR
SCF16 - 27%
SCF19 - IDR avg. Loan Balance $62k
SCF16 - $43k
SCF19 - IDR enrolled BA only 31%
SCF16 - 26%
SCF19 - IDR enrolled Household Wages Avg $70k
SCF16 - $62k
The Collier models using the '19 data have higher R2 and more loan debt variables flag as significant - mimicking trends found in Collier (2020) but not our former paper using '16 data.
Consistently, <$12.5k households remained less likely to be enrolled in IDR, as are $100K+
Consistently, <$12.5k households remained less likely to be enrolled in IDR, as are $100K+
We saw that households earning $55-75K were more likely enrolled (than $40-55K) - we believe was due to the generous terms of REPAYE.
In these models, we lost the consistent significance for female and minority borrowers when applying the models to the '19 data.
In these models, we lost the consistent significance for female and minority borrowers when applying the models to the '19 data.
The Looney/Yannelis models captured increased female and minority borrower enrollment with the '19 dataset - previously did not see differences for minority enrollment with '16.
The previously consistent finding of having $50k+ in student loan debt was non-significant here...
The previously consistent finding of having $50k+ in student loan debt was non-significant here...
Given Collier models, we lowered the floor in the L/Y models to $40k, found significance again.
Therefore, we believe this outcome suggests REPAYE has lowered the floor for entry into IDR and that people either WANTED or NEEDED the financial safety IDR provides.
Therefore, we believe this outcome suggests REPAYE has lowered the floor for entry into IDR and that people either WANTED or NEEDED the financial safety IDR provides.
Also looked at financial differences between those in IDR v. Traditional Repayment. Pic 1 shows significant differences for lower amounts in traditional checking accounts (-$1,004) and a lowered chance of saving for retirement (-7pp).
Pic 2 shows null findings with '16 data.
Pic 2 shows null findings with '16 data.
So what does this all mean?
Well... AGAIN we could not bring clarity to the conversation we were hoping to bring BUT this shows that how you go about generating models really, really, really matters to outcomes. So please be careful here and run many models.
Well... AGAIN we could not bring clarity to the conversation we were hoping to bring BUT this shows that how you go about generating models really, really, really matters to outcomes. So please be careful here and run many models.
We argue that given this lack of clarity to look for the trends...
In that female and minority, borrowers are more likely to be in IDR. Those with higher balances remain more likely enrolled in IDR.
In that female and minority, borrowers are more likely to be in IDR. Those with higher balances remain more likely enrolled in IDR.
Those with the lowest household earnings are less likely to be in IDR - which is curious and we should go find out why. However, the "savvy" highest earning people (and those with professional degrees, maybe) are less likely.
Many financially related outcomes remain statistically similar (homeownership) - but some trends seem going a different direction for some outcomes like retirement participation/savings.
ATM, IDR is still providing some measures of equalization for those with higher balances.
ATM, IDR is still providing some measures of equalization for those with higher balances.
Finally, REPAYE has seemingly lowered the barrier for entry when determining debt loads and opened access for more households in an earnings bin that encapsulates median household income ($69k)
https://www.census.gov/library/publications/2020/demo/p60-270.html#:~:text=Median%20household%20income%20was%20%2468%2C703,and%20Table%20A%2D1).
https://www.census.gov/library/publications/2020/demo/p60-270.html#:~:text=Median%20household%20income%20was%20%2468%2C703,and%20Table%20A%2D1).
P.S - I cannot tell if the tables are gonna come out blurry. I hate how the desktop version of Twitter "crops" things. So, download the pieces.
