Is gentrification a cause?

I recently listened to this UCLA Housing Voices podcast, “Geographies of Gentrification with Hyojung Lee”. It was based on this paper.

From the paper:

The large and growing body of gentrification research has not yielded consensus on whether or not gentrification causes residential displacement. One explanation for inconsistent findings across studies could be different definitions of gentrification used in the analyses . . . Most of these studies, however, report that their results are robust to different measures of gentrification.

A more compelling explanation for the inconsistency between single-city and national-level studies is that the association between gentrification and residential mobility differs between different types, or clusters, of metropolitan areas.

The authors attempted to get more detail about the effect of gentrification in different types of cities. Basically, the idea is that maybe different studies have produced different results because they are feeling different parts of the elephant.

I wonder, however, if the problem lies with treating gentrification as a causal process at all. It might make more sense to see it as a process that is the result of other causal factors. It is more of an accounting identity than a cause.

My price/income approach may help thinking about this. Just to review quickly: American cities have developed a pattern where cities with higher median incomes tend to have housing costs that claim more of local incomes. (They have a higher median price/income or rent/income level.)

Figure 1

This is really weird. Within any given city, richer households spend a smaller portion of their incomes on housing. This relationship didn’t really exist between cities until recently. The relationship is likely due to housing supply constraints. Where homes can’t be built, families with lower incomes are forced out. Much of the reason some cities have higher median incomes is because many poorer residents had to move away because there wasn’t enough housing.

You can see this more clearly within those cities. Figure 1 is reproduced in Figure 2. Figure 2 also includes ZIP codes within Atlanta and Los Angeles. You can see that the median price/income ratio within a given metro area is really a product of the price/income ratio in the ZIP codes with the lowest incomes. And, if the median is picking up a signal of higher prices within a metro area, it is just because it is picking up half of the rise in ZIP codes with lower incomes.

Figure 2

These patterns, which are relatively universal among the large metro areas, give insight into the issue of “gentrification”. Maybe Figure 3, from my paper, helps with this.

Figure 3

The left panel is an idealized picture of Atlanta from Figure 2. There are all sorts of changes in a growing city - actually in any city. Many neighborhoods are relatively stable. Some neighborhoods are growing. Some, especially in cities with stagnant economies or populations, can decline or accumulate vacancies. Some change because of new developments. Some change because the average income of the neighborhood population increases or decreases. Maybe, as a city grows, the convenience of a certain neighborhood to various changing metropolitan amenities starts to draw a different population. Maybe a formerly sleepy neighborhood starts to attract affluent new households.

In this mess of changes, the change that involves new development associated with rising incomes within the neighborhood is called “gentrification”. Basically, in the left panel, gentrification is the blue lines pointing to the left. Because of reasonable worries about the effect of change and potential displacement on families of lesser financial means, this particular part of the boiling urban stew has a name and an associated concern.

In that idealized mess of arrows in Atlanta, does it make sense to say that gentrification “caused” anything? I think it makes more sense to recognize that a living city must incorporate change. We call some of that change “gentrification”. Those types of changes create certain types of stresses and opportunities that differ from other types of change. In that way, I suppose you could say that the type of change we call “gentrification” “causes” certain outcomes. But, I think, the reason that language is unhelpful, is that you can’t just erase those leftward pointing blue arrows in the left panel of Figure 3, and leave all the other changes in the city intact. All those changes are part of a larger stew.

Now, let’s move to the right panel of Figure 3. One thing that happens in a city that doesn’t build enough is that the mess of changes becomes increasingly directional. In Atlanta, there was a tangle of substitutions across the metropolitan area, as families constantly make decisions on the margin to match their location with their idiosyncratic preferences and their personal station within the socio-economic hierarchy of the city. In a city like LA, increasingly, all the substitutions are in one direction.

In Atlanta, the question was “what neighborhood suits us best?” In LA, the question is, “what neighborhood can we afford?”. More specifically, since being located in LA is associated with a large cost premium, the question is “where can we outbid households of lesser means in order to make the compromises we need to make to offset the LA cost premium?”

Every housing change in LA, then, becomes more gentrification-like. “Gentrification” isn’t “causing” displacement in LA. The lack of housing in LA is causing more LA housing decisions to fit the description of “gentrification”.

Does more public housing reduce gentrification?

Lee is very careful about making policy claims. On the podcast, the one policy suggestion that Lee alluded to was to build more public housing (around the 1:07:00 mark). But, I think the point of that policy suggestion is to try to do the thing I described above - it’s a way to try to erase the leftward pointing blue arrows in figure 3. The point of proposing public housing here is to imagine that it is exempt from market forces.

Those sorts of policy solutions basically rearrange the dots on the scatterplot in Figure 2, but they don’t do anything to flatten the slope. The pressures of what we call gentrification, especially the problematic pressures, come from the steep slope. This is why I think the price/income slope can help to sort out these causality issues. Most of the time, measuring the effects of interventions is an exercise in what you have accounted for and what you haven’t. When looking at any local intervention, it’s a lot easier to measure the idiosyncratic movements of a few of those dots than it is to estimate the effect on the slope of the line. So, to my mind, the most important aspect of practically all of the attempts to quantify the “effects of gentrification” is the amorphous blob of unmeasured, unseen data that exist outside the data set.

Most research on local effects of housing policies and interventions measure idiosyncratic changes in the dots in Figure 2, but the solution to housing costs must be to flatten the slope of the line in Figure 2. Making this distinction is key to measuring the right things.

Fixed Effects

On that topic, I think there may be issues with using regional fixed effects here that are similar to what I have discussed about housing bubble research. In that research, the differences between LA and Atlanta are controlled for with “fixed effects”, except that as far as I can tell, nobody has noticed or accounted for the main difference between LA and Atlanta, which is the difference in that slope. So, there are two problems with controlling for fixed effects. (1) If you’re analyzing home prices, the difference between LA and Atlanta is 90% of the story. and (2) In the process of removing 90% of the story, if it is being removed in a way that doesn’t symmetrically account for the differences, the adjustment itself is bound to create a bunch of biases that are at least as important as the 10% of the story you have elected to analyze.

In the gentrification paper, the expensive coastal cities don’t particularly stand out as having particularly bad effects from gentrification. But, the authors controlled for fixed effects. I think this limits their ability to see the heterogenous issues they are looking for. The effects of gentrification are going to be scaled to the slope of that price/income line. They are scaled to the housing scarcity associated with the entire metropolitan area. Depending on exactly what dimensions you choose to control away the differences, surely some of the differences in effects will dissipate with those controls.

I suspect that their results are very different when they don’t use MSA fixed effects. Also, even the definition of gentrification is affected by this. Part of the paper’s definition of gentrifying neighborhoods is “Among the neighborhoods at risk of gentrification, we classify a tract as gentrifying when the neighborhood experiences a substantial increase in its within-MSA percentile rank of college graduate share over the time period examined.” So, the systematic directional housing activity in the right panel of Figure 3 is controlled away. The most gripping problem with gentrification is mostly that we have entire metro areas that are gentrifying. Relative changes within metro areas can’t see that.

This seems similar to the housing bubble research. Since the most problematic displacement associated with gentrification comes from those systematic, metro-area wide pressures, the most important details are swept away in fixed effects.

Changes Since the Great Recession

Furthermore, as I have written about extensively here (and have a paper due to be posted any day, specifically, on the topic), every metro area is becoming more LA-like. Since the Great Recession, every major metro area looks less like the left panel of Figure 3 and more like the right panel.

The universality of the housing supply problem is making it harder to find distinctions among different regions. This is creating an odd problem. It has been appropriate to treat these problems as local problems, but the more those problems spread everywhere, the less we can learn by studying the differences between areas. And, the inability to see those differences might actually lead to an inappropriate sense that the problems are less acute, or that they aren’t in fact getting worse.

In summary, I would like to see research like this poking at the very inclination that gentrification should be treated as a causal factor rather than working harder at measuring its causal importance. I think more wisdom will come from that sort of conceptual framing.