Gentrification in Charlottesville: “We Ain’t Seen Nothing Yet”?

  • Previous research has shown that Charlottesville has not undergone a deteriorating long-term trend in housing affordability.
  • Wage data shows wages keeping pace with and for some stretches exceeding housing price growth.
  • In our research we note that one way this data might be misleading is if wages are keeping pace not due to the increase in wages for existing residents, but via a change in the composition of residents.
  • Publicly available data suffers from limitations, particularly in a geography with a large student population, that makes it difficult to measure displacement.
  • We exploit Quarterly Workforce Indicator data at the MSA level to get around some of these limitations to test whether for evidence of labor force cycling that brings in high-wage earners and leads to exit of lower-wage earners.
  • QWI data for Charlottesville shows a similar pattern to MSAs with clear record of recent high-wage earner exits and stark differences from “Zoom-town” MSAs that have a clear record of high-wage earner influx.

CFRP has released several pieces of research that looked at the question of housing affordability in Charlottesville.  The pro-upzoning narrative relies in part on the assertion that Charlottesville has experienced an unusual increase in housing prices, and that this increase is not merely a function of general inflation (of housing production costs and also resident wages).  We have spent a great deal of time and effort trying to find evidence of such a phenomenon, but our analysis shows pretty clearly that Charlottesville is if anything a laggard in housing cost increases.  In fact, wage growth relative to housing price growth in Charlottesville has been quite lot better than the US average.  From 2010-present we are at the 41%ile of counties in terms of housing price increase (using Zillow data), but over 90%ile on wage increase (BLS) and median income increase (ACS).  In outright levels, we are at 60%ile for housing prices and 73%ile for wages.

The potential hole in our analysis, though, is that we have only been able to measure changes in wages and income at a fairly high level of aggregation.  At the aggregate level, wages and income can increase because the pre-change population of workers manages to notch wage gains; or they can increase because the pre-change population of workers is not the same people as the post-change population.  That is, an influx of high-wage workers can coincide with the exit of lower-wage workers (assuming low overall population growth, as in the case of Cville City, or low-moderate growth as in the case of the MSA as a whole).  In other words: displacement and gentrification.

So, which is it?  Housing prices and wages rising together as part of general process of price and wage inflation involving mostly the same or similarly situated people?  Or housing price affordability as a product of an increase in wage / income growth that is endogenous to the housing market – housing staying affordable by definition as those who earn less and can’t afford it leave?

The rest of this write-up is structured as follows.  First, we will go over why the standard publicly available mobility data could not answer our question.  Second, we will describe the QWI data that we use for this analysis and why we believe it gets at the question we are asking.  Third, we will describe some shortcomings of the data.  Finally, we will share the results of our analysis of the QWI data.

The Data That Didn’t Work

Our first instinct was to look at Census county-to-county population flows with Charlottesville as either source or destination.  We would look for net flows in from known “high-wage/high-opportunity area” and flows out to known “lower-wage/lower-opportunity areas.”   Several problems, though.  First, this is a 5-year-ACS-based sample, so it is highly lagged.  Second, even with the five-year window, counts are so small that the margin-of-error for any Cville/Other County flow is typically larger than the flow itself.  Third, there is no ability to filter, so this likely captures a lot of student flows (and these will be distorting, as we will see heavy inflows from other VA counties and outflows to NY, DC and other young-professional places; which would mask on a net basis the “gentrifying” flows from those places).  Fourth, income could only be measured by proxy — for example, looking at the income medians for source and destination counties.  But that is unsatisfactory, because each of these counties has highly heterogenous income levels and movers are not a random sample of county population.  In sum, the county-to-county flows data will not work.

We then looked at the Census aggregation of the migration data.  The ACS does ask people “where did you live last year?”  Several variables are cross-tabbed with response categories on mobility (same house / same county/ same state / other state) — including income categories.  Unfortunately, this will only tell us about people who move to Charlottesville.  It would be great if Census tracked people at their destination (i.e. compiled statistics for people who left), but it doesn’t.  In addition, we have the same problem with students.  Many of the “same state” category are students, who show up and have very low income.  According to this data source, the incomes of “movers-in” is far lower than that of people who were here a year ago.  But we know that is almost certainly not the case for the population we care about (non-student households).

Finally, we looked at the IRS’s migration stats.  This gives you very good data on incomes of people who stay and people who go from a jurisdiction.  It has relatively high granularity with respect to time, reporting annual data.  But there is a rub here, too.  The IRS only reports the “Year Two” income.  If people are moving and having their income change due to the move, you will get a big distortion.  Unfortunately, this perfectly describes graduating students.  The median starting salary for UVA grads reaches something like $70K!  So when we look at these numbers, we see veryhigh incomes of leavers.  But one cannot conclude from that that Cville is losing rich people.  Most students coming to UVA don’t have tax returns of their own before coming (undergrads, anyway), so the data for “movers-in” is not as distorted.  Still, if we cannot rely on the “move-out” numbers, the data source is not useful.  Finally, the IRS data is mean AGI.  That can be highly distorted by upward outliers, and it includes all income, not just wages.  Very high-income returns tend to show a lot of non-wage income.  One very wealthy business owner with a $10mm AGI will massively skew the numbers.  We can probably safely use mean wage numbers when medians are not available, but mean AGI is a different story.   It was interesting to look at this just for flows between Cville City and the other counties in the MSA, though.  What we found is that there was not a big difference between the AGI of people leaving Cville for Greene, Louisa, or Fluvanna, vs people moving in from those places, nor actually a large net flow.  That wouldn’t tell you about the ZoomTown effect, but would suggest that the story of “poor people being pushed out to the periphery” is not necessarily straightforward truth.  We have some details on that in our commute analysis.

Quarterly Workforce Indicators

The Census publishes quarterly numbers on jobs flows as part of their Quarterly Workforce Indicators.  This is an administrative dataset, not a survey dataset.  It is derived from state tax and unemployment insurance filings.  Job flows are broken down by Metropolitan Statistical Area of source and destination.  The data also can be filtered for job transitions that started and ended in stable employment.  That filters out almost all students.  We can also filter by age, which we used as a check on distortions of the data from student flows.  We look at job flow counts and average earnings for within-MSA transitions and then MSA-to-everywhere-else transitions (and same for from-MSA), and from that we can derive the earnings results for transitions to and from Charlottesville to “rest of the country”.  We can do the same for some other cities where we have clear evidence of high-earner influx or high-earner flight, and then compare Charlottesville’s numbers to those. QWI is not without shortcomings.  The wage values are average rather than median.  As discussed above, though, the right tail for wages is less extreme than for tax-return income.  The data is reported by MSA and not by county, so we do not learn much about transitions from the city to the surrounding counties.  That said, no one has argued that gentrification is a product of people moving into the city from the surrounding counties.  We will, however, miss some of the theorized flow out of the city.  Note, however, that “over the mountain” is not part of the Charlottesville MSA (Staunton, Waynesboro and Harrisonburg areas are not part of our MSA), so we catch those moves, if they involve a job move.  Someone who moves over the mountain, but remains working at the same job in Charlottesville will not be captured.  We do, though, have as a check the IRS data mentioned above, and we reiterate that this data shows no dramatic net moves of low-earnings returns from the city to the surrounding areas.  While some remote workers might be missed by QWI if they remain on the books of their source geographical office, the “convenience of the employer” rule gives a strong incentive to many remote workers to administratively establish their place of employment in their destination.   Finally, since QWI tracks job moves, we will miss out on flows of retirees and of people out of the labor force more generally.  Clearly, some amount of gentrification could be driven by wealthy retirees moving in and poorer retirees moving out.  Likewise, if people leaving were those who suffered a long-term disconnection from the labor market, we would miss those flows as well.  We would note, though, that for the purposes of forming zoning policy, it is not likely that upzoning would have any effect on the production of housing affordable to people not in the labor force.  That market segment requires heavy subsidies, not market production.  Overall we believe the QWI data can tell us a lot about the validity of one narrative of gentrification: namely, are high-earners from Northern VA, New York, and other high-opportunity areas moving to Charlottesville and are we seeing clearly lower earnings among movers-out? 


We ran parallel analyses for Charlottesville, a couple of MSAs (Chicago and San Francisco) where there is strong evidence of a trend of high-earners moving out and second group of MSAs (Tampa, Boise, and Austin) where there is strong evidence of a trend of gentrifying inflows.  We looked at the following ratios:

First, we looked at the ratio of wages of people who move into a target MSA post-move to the wages post-move of people who move out of the target MSA.  We filter only for moves from stable employment to stable employment.  Typically, people moving jobs have higher income in the job they switched to than in the job they left.  This ratio should control for that by looking at post-move vs post-move earnings.  However, it can be distorted if we are looking at a particularly high-cost or low-cost MSA.  A change in wage may largely reflect a change in relative cost of living.  Still, it is a good start. 

Second, we looked at this ratio, except that we compared the post­-move earnings of jobholders into an MSA to the pre-move earnings of job switchers moving out if the MSA.  Since job switching is typically associated with an earnings jump, this measure should be biased upward.  But since we are comparing earnings from the same MSA in this case, we get around the cost-of-living distortion.

Third, we looked at the ratio of earnings between new arrivals and those in the MSA who never changed jobs at all.  This metric should be higher for cities undergoing a gentrifying influx.  However, the overall level of this metric tends to be biased downward because job-switchers tend to have lower earnings.  They are often earlier in their career than stayers, who are more established.  Still, since this bias is present in all MSAs, we believe the ratio has information value.

Finally, we look at to what degree people moving in (and people moving out) of an MSA saw an increase in earnings from their old job to their new job.  That is, we look at the same people and compare pre- and post-move earnings.  Did they move for a raise?  This ratio would tend to detect if an MSA is high or low cost, and also if there is a sudden productivity shock that allows for new high-wage jobs where the wage is an attractor.

Let’s look at each of these metrics, one-by-one.  First, post-move earnings of movers-in and movers-out. 

Charlottesville well off the “ZoomTown” cities’ ratios

Charlottesville shows a consistent trend since 2019 of lower earnings for movers-in than for movers-out.  The level is similar to the post-pandemic ratios for Chicago and San Francisco.  We very clearly see the pandemic move-out effect in the San Francisco data, which gives us confidence that we are picking up real trends.  On the other hand, Austin, Tampa and Boise all show ratios well over 100%.  Movers-in earn more than movers-out.  And here, too, we see the spike in the middle of the pandemic.  In the cities with strong long-distance gentrification flows, we see those flows in the data, providing more confirmation that our chosen data source sheds real light on the question we want to answer.  Our first metric therefore casts serious doubt on the narrative of extreme gentrification flow in Charlottesville.

Next, the results for the ratio of earnings of movers-in in their new job versus the earnings of movers-out in their old jobs.  Recall this metric will be biased upward for all cities.  However, it should capture a situation where people are being forced out and responding by finding higher paying jobs somewhere else as a response.

The ratio is higher in all cities, as predicted.  Once again, San Francisco and Chicago come out well below Tampa, Boise, and Austin.  Charlottesville falls in the middle on this metric, slightly above SF and Chicago and well below Tampa, Boise and Austin. The gap between the first metric (post-move earnings vs post-moving earnings) and this metric is larger for Charlottesville than for the other cities.   We think this may relate to the fact that Charlottesville is still a low-cost area, relatively speaking. That is supported by the fact that the other cities with a large gap, Boise and Tampa, are also lower-cost areas (though increasing in cost).  The weighted average price level of destinations for people leaving our area for a new job may just be higher than Charlottesville’s.  We will see more evidence of that below.  But again, this metric shows a clear distinction between Charlottesville and the gentrifying city group.

We turn now to the ratio between the earnings of movers-in and the earnings of people who did not change jobs.  Are people who move in earning more than stable local job-holders?  We would expect this metric to be biased downward for all cities, since young and early-career workers have more job mobility than more established workers.

Here we see Charlottesville toward the bottom again, basically in line with Chicago and SF.  Austin comes in lower on this metric than the others (even with Charlottesville), possibly because it had a larger pre-existing base of very high-wage jobs at the beginning of our sample.  Tampa and Boise are quite a bit higher.  Again, no evidence in Charlottesville that new arrivals are blowing established job-holders out of the water on earnings.

Finally, we consider the question of to what degree moving in or our of a jurisdiction is associated with an increase in earnings.  As we have mentioned above, generally job-switching (for people moving from stable employment to stable employment) is associated with an increase in earnings, particularly in a tight national labor market.  So both move-ins and move-outs usually show an increase in earnings on average.  We focus on the relative “raise” related to moving into an MSA vs moving out.  We believe this measures broadly two factors.  First, how high is the relative cost-of-living of the MSA.  People will not move to a high-cost MSA without a raise; people may not demand a raise if they are moving to a lower-cost MSA.  Second, has there been a productivity shock in the MSA such that a sector is suddenly pushing up the wage scale.  An example of that would be North Dakota during the shale boom. 

On this metric Charlottesville is the clear laggard.  Boise and Tampa actually come out below zero, which we would attribute to the fact that they still have much lower cost-of-living than some of the high-opportunity MSAs they are drawing people in from, for lifestyle reasons.   Chicago and SF, while they show de-gentrification on other metrics, come in right around zero on this metric, probably because they remain high cost-of-living MSAs.  Charlottesville shows no sign, therefore, of being either a high-cost-of-living MSA or in the throes of a wage-bending productivity shock. 

While QWI data has some notable shortcomings, we believe it is fair to expect if an area is in the throes of sharp gentrification and displacement, it will show up to some meaningful degree in that data. We can certainly detect it in the data for well-known cases of gentrification and high-wage in-migration like Tampa, Boise and Austin. Likewise, we detect the reverse in notable “outflow” stories like Chicago and San Francisco. We simply see no evidence of what the upzoning narrative confidently pronounces on the basis of nothing more than anecdote, namely strong gentrifying displacement. Yet another case of activists selling an unsophisticated Council and a well-meaning populace on a just-so story in order to push through the very prescription that happens to align with their imported land-use-regulation ideology. If people don’t like housing prices now, before we have had truly seismic gentrification flows, they will really hate it if unfettered development of luxury housing turns on the sort of gentrification tread-mill in evidence in places like Austin. Due to the limitations of our data, we would not claim that this analysis is anything like the last word on displacement. But it is highly peculiar that the city’s consultants have not produced anything at all close to it in terms of analytical rigor. Perhaps a good question to ask at the next pop-up is why they haven’t…

March 2023

Addendum: IRS Raw Inflows

As we mentioned above, IRS tax return data has some serious shortcomings that prevent it from giving an accurate picture of displacement. However, one quantity that is still interesting to look at on its own is the raw number of tax filers that move each year from selected jurisdictions to Cville. The inflow numbers should show less distortion from students than the outflow numbers — recall the outflow numbers can be highly distorted by graduating UVA professional students. And if we look only at raw counts of returns rather than income, we don’t need to worry about the distortion to AGI caused by a few ultra-high earners. What the inflow numbers give us is a sense of the absolute maximum size of flows from jurisdictions of interest. We need to bear in mind, though, that gross inflows will surely overestimate migration impacts, since some number of non-students do move out of Cville to these selected jurisdictions. But if the gross inflows are not large, that can rule out the proposition that Cville is in the grip of a mass-migration from wealthy jurisdictions. It can also illustrate whether there is a strong increasing (or decreasing) trend in gross migration.

So what did we find? We looked at the IRS data for moves from New York City, Washington DC, and Northern Virginia (Arlington, Fairfax, and Alexandria). The number of tax filers moving into Cville City from those areas has remained rather modest from 2013 until the latest data (just released in April 2023) for 2021. We are covering the peak “COVID migration” period. We see very little evidence of either high levels or of an increasing trend. The total inflow from these jursidictions has amounted to around 200-250 tax returns per year, in a city that has approximately 17,000 resident tax returns filed. And remember this is gross inflow. It does not net out non-student out-migrants nor does it net out inflows of students (some number of whom, particularly among graduate students, will have filed tax returns both before and after their move to Cville). You can see the data in the chart below. Another piece of hard evidence refuting the Upzoning Narrative.