-->

Wednesday, July 8, 2015

Landscapes of Predation: The Value in Poverty and Racism

A few weeks ago Sarah Kendzior wrote a piece in the Guardian on pay day loan establishments in Missouri. It stood out to me because not only is the issue of predatory lending an incredibly important, and criminally under-reported phenomenon, but that it reminds us that while poverty is indeed expensive (an idea that has been relayed constantly here, here, and here are some examples) it is less common for journalists to call out the uneven geography of predatory servicers or to point out that many, many people make a lot of money off of the poor. So, if poverty is expensive then it stands to reason that someone is pocketing those expenses. But the other half of that equation is rarely discussed. 

Kendzior's piece explores this dynamic by pointing to the many people trying to regulate payday lenders and seek restitution for exploited borrowers, but, for once, I once wished that a news article had more "data". She offers a series of pronouncements about payday lenders in the state that, to me, were screaming for some better data visualization and exploration or, at minimum, some sources. For example, Kendzior writes there are 958 more payday loan establishments than there are McDonalds in the state of Missouri. This number is a variation of some earlier numbers reporting more payday loan establishments nationally than there are McDonalds based on a 2014 St. Louis Federal Reserve report. It is not that I disbelieve payday loans are more numerous, it wouldn't surprise me at all. But nearly 1,000 more establishments in one state seems extreme and some source for the claim would be useful. It should also be noted that figure does not actually conform to what the state's survey of payday lenders says. Further, the author quotes reps from the state's Dept. of Finance on the targeting of poor populations in the state. But what does this actually look like and what are some indicators of the severity of the differences between the poor and non-poor, and in Missouri, the black and non-black?

In an attempt to try and answer these questions I downloaded the Starbuck's global map of 2013 from Socrata and I used  Reference USA to grab every McDonalds in the state of Missouri (standard caveats apply as to the imperfect nature of such listings, but I am confident I have the vast majority of McDonalds in the state). The state of Missouri provides a list of all registered payday lenders in the state. I then geocoded the McDonalds and payday locations. Finally, I grabbed some of the most recent 5-year ACS data at the census tract level for the state of Missouri and joined everything together in order to start to get an idea of the uneven geography of predation.

What follows are some preliminary results. I don't pretend to have all of the answers here, but I hope this little exploration can provide some clarification and to, ideally, make us all consider the differentiated social and economic landscapes we travel through every day.

Finding 1- Yes, there are more licensed payday lenders than there are McDonalds and Starbucks combined

 Yes, there are more payday lenders than there are McDonalds and Starbucks combined, according to my analysis. By my estimates, Missouri had 799 licensed payday lenders (I lost some due to bad addresses in geocoding) and a little less than 500 Starbucks and McDonalds.


Total Establishments
Payday 799
McDonald's 334
Starbucks (2013) 156

By these estimates there are a third more payday lenders than MickeyDs and Starbucks with a combined difference of 309 establishments. Now, this is a still staggering number, but 958 is flat-out wrong if Kendzior is citing Missouri's numbers for payday lenders.

Finding 2- Yes, payday lenders, on average, target poorer neighborhoods (census tracts)

Unsurprisingly, payday lenders tend to congregate in poorer census tracts. I created a dummy variable for the presence or absence of a payday loan location and looked at the poverty rate difference between the two. For those census tracts with a payday location their median poverty rate (I divided the census count of people in poverty by the number households they could determine poverty status for) was approximately 18% while for tracts without payday locations the median poverty rate was 14%.

An alternative way of visualizing this difference is a good old fashioned box plot where we can see a clear difference in the median poverty rate between tracts that do and do not have payday loan locations. 


While the basic numbers and the boxplot show there is some difference between these two types of tracts is the difference statistically significant? One way to answer this is to use the Kruskal-Wallis test. Not to get into too much depth but it's a way to test to see if the medians of some set of groups is different when their underlying data is not normally distributed. So, I ran this test and the poverty percentages between these two types of tracts was highly significant with a chi-square value of 42.12 and p < .001.

Finding 3- PayDay Lenders ARE NOT targeting the most Black census tracts

This finding is the most interesting to me because it forces us to think about a spectrum of Black neighborhoods, something that too often is ignored in a lot of writing. Running the same tests as before I found no difference between the census tracts with and without payday loan locations in terms of the percent of the population that is Black.


How can this be? My own theory is that the blackest census tracts are potentially too poor to be attractive targets for payday lenders.

Let's take another look at the data. The following plot is a scatter plot of census tract's percent white population and poverty share. There is a clear negative and significant relationship between the percent of a census tract that is white and poverty rates. Now, this is not to say that there are no white census tracts that are not poor. The figure places a lie to that assumption, but it is clear that there are a vast number of a census tracts that are almost entirely white with very little poverty.


Conversely, Black census tracts are overwhelmingly poor and get even poorer the Blacker the population. Notice this is nearly a mirror image of the plot above and a testament to the long lasting effects of entrenched racism and segregation within Missouri. But we should notice that there are many, many tracts with up to a quarter of their population identifying as Black that do no have terrible amounts of poverty. I don't want to belabor this point, but these figures are to show that there is a lot of variation with respects to poverty when it comes to Black census tracts even though the direction of the association between the two is terrible.

But what does this have to do with payday loan locations? Payday establishments, like any business, even predatory ones, still require a viable customer base. And payday loan locations seem to show a clear preference for poor, but not too poor, areas. In other words, payday loan establishments seek to situate themselves in working class, or working poor, neighborhoods as opposed to the most destitute. This makes intuitive sense. Even within its own name we get a clue. Payday loans still require some guarantee of repayment and offer their services often times as a form of emergency bridge funding for precarious workers. So an optimal payday location is in an area where a lot of people are working, but they are working in low wage/precarious industries, but still make enough money to be worth the trouble.

As the figures above show, the blackest census tracts are some of the most impoverished census tracts in the state. Two following figures give a way to visualize this.



This figure shows the relationship between payday loan locations and census tract poverty for all census tracts. The fit line shows a fairly smooth curve that shows an increasing relation between poverty and the frequency of payday loan locations and then a decreasing tendency as poverty increases.

In order to get an even clearer picture let's focus only on tracts that have payday loan locations. Here the relationship is even clearer, though we see there is still an increase in the frequency of payday loan locations until leveling off and dropping at around a 40% poverty rate. Now, a 40% poverty rate is still tremendously high, but I would draw your eye to the more extreme values on the y-axis and we can see that where people are most exposed to payday loan locations are in less poor tracts.

Finally, what is the actual strength of these associations? To try and answer this I ran three logistic regressions where I regressed the presence or absence of a payday loan location within a census tract against the share of the tract population that's black, that is in poverty, and an interaction term that is the multiple of the percent black of the population and a dummy variable for the median value of the tract population in poverty. The interaction term, because 1 is for all tract with a poverty rate greater than the median (approx. 14.6%) should give us the relationship between the odds of a payday location being in a particular tract that is both black and low income.


Logistic Regression Results for Pay Day Locations
Dependent variable:
pd_dummy
(1)(2)(3)
blackshare-0.170-0.951***0.004
(0.240)(0.279)(0.801)
pov_share3.257***1.930**
(0.553)(0.781)
pov_dummy0.511***
(0.184)
blackshare:pov_dummy-0.990
(0.850)
Constant-0.848***-1.314***-1.376***
(0.067)(0.107)(0.121)
Observations1,3931,3881,388
Log Likelihood-844.750-825.247-821.290
Akaike Inf. Crit.1,693.5001,656.4931,652.580
Note:*p<0 .1="" sup="">**




These three models repeat what the figures ahead already told us, but now we have an idea about the direction and strength of some of these variables. First, note that the percent black variable is quite unstable jumping from significance back to non-significance and even changing signs. But take a look at the poverty share variables. In terms of more easily understandable odds ratios, for every unit change in poverty a census tract is nearly 7 times more likely to have a payday loan location.

We know from our graphs that this effect is moderated at the most extreme ends of impoverished neighborhoods but there is a clear strategy here.

Where to go from here?

These tests are by no means definitive, but they do challenge us to think deeply about the different ways that the landscapes we navigate every day are constructed in drastically different ways depending on who lives where. Missouri also shows us just how perverse racial segregation is. We are all now well aware of a smaller kleptocracy like Ferguson that is seemingly built upon subjugating its black residents, but it also shows up in how we have to think about something like payday loan locations.

Blacks are incredibly concentrated in the state, isolated in three major cities/metro areas. Not only are they isolated in those cities but even within those cities racial dividing lines are stark. As such, it becomes easy for people, including social scientists and commentators, to sloppily lump all Black people together as living in one giant, unmitigated ghetto when that is clearly not the case. There are many census tracts with sizable Black populations that are not incredibly poor, but it is an absolute fact that the blackest census tracts are all poor whereas the whitest tracts in the states are the least poor.

This is not to relieve payday lenders of their responsibility in targeting Black neighborhoods but we potentially miss the mark if we see their strategy as solely, or even predominately, as a racial one and not a class/income based one. The problem here is poverty. It is just that in Missouri, and the rest of the US, poverty remains so terribly racialized and Black peoples are still widely seen as uniformly poor and culturally deficient that we cannot imagine landscapes where one's Blackness is not the principal reason why some external institution would seek to exploit you.

Bonus!!

If you managed to make it thus far, thank you. I truly appreciate it. But as a minor reward here is a map of the payday loan locations I made. I hope it can prove useful.




No comments: