This section summarizes our methodology for mapping Austin’s neighborhoods as either gentrifying or not gentrifying, and for classifying the gentrifying neighborhoods according to their stage of gentrification. Our procedure is an adaptation of a method devised by Dr. Lisa Bates of Portland State University in Oregon, and first applied to Portland.23 Note that this section provides a high-level overview; full methodological details and a step-by-step procedure can be found in Appendix 3.
What we analyzed
The basic geographic unit that we used to analyze Austin is the census tract. A census tract is an area defined by the federal government that typically contains between about 2,500 to 8,000 people. It can be thought of as roughly equivalent to a neighborhood, although census tract boundaries do not necessarily line up with neighborhood definitions commonly used in Austin. The geographic size of a tract depends on how many people it contains and how densely populated it is. As one example, Travis County census tract #15.04, which covers the Crestview neighborhood in North Central Austin, is just over a square mile in size.
We began by identifying all of the census tracts that lie either entirely or partially inside Austin’s city limits. Next, we eliminated several tracts from the study because they are unusual places not subject to the typical processes of neighborhood change. These included the tracts containing Austin Bergstrom International Airport, the University of Texas (UT) main campus, and Camp Mabry, a military base. We also eliminated two tracts comprising the West Campus neighborhood immediately west of the UT main campus, since demographic information from student-dominated neighborhoods can lead to misleading conclusions. For instance, a very high proportion of college and graduate students are represented in official data as living in poverty, even though many of them have access to opportunities and resources that are a world away from what predominates in a truly impoverished neighborhood.
After we had eliminated several census tracts from the study, we were left with 200 of them. We assigned names to each of them, which appear on the interactive map interface that we have released alongside this report. The names represent our best attempt to match the various census tracts in Austin with locally meaningful geographic descriptors.
It should be noted that the Census changes its definitions of tracts every ten years (following the release of each new decennial census). We used Census-provided “crosswalk files” to harmonize the boundaries of 1990 and 2000 tracts with 2010 tracts.
Overall procedure
Following the Bates methodology, our analysis unfolded in three steps. The ultimate goal behind the procedure was to classify every census tract in Austin as gentrifying or not, and to classify the gentrifying tracts into five categories based on the following stages:
To get to this classification of gentrifying neighborhoods, our first step was to classify each census tract on the basis of vulnerability. In general, vulnerability refers to a tract having an above-average share of vulnerable residents—classes of persons who are more likely to be displaced when housing costs rise in an area or an area is subject to increased public and private investment (see the above section for a more detailed description on vulnerability). Each tract was classified as either vulnerable or not vulnerable. The second step was to classify tracts based on demographic change: Between the years 2000 and 2016, had the census tract experienced an increased share of residents associated with gentrification (e.g., white, higher-income, highly-educated, homeowner residents)?
Finally, the third step examined housing market change from 1990 to 2016 and from 2000 to 2016. For this step, census tracts were classified according to whether they had experienced an above average amount of appreciation since either 1990 or 2000, or whether they were adjacent to a tract that had experienced such change (typically an indication, according to research, that home price appreciation will soon take place).
After collecting this data, we assigned one of the five gentrification stages to each gentrifying neighborhood. A neighborhood was classified as an Early: Type 2, Dynamic, or Late stage gentrifying neighborhood, if the census tract met all three of the following conditions: (1) an above- average share of vulnerable residents, (2) experienced significant demographic change, and (3) experienced significant housing market change. If a tract was vulnerable and had experienced appreciation but not yet demographic change, it was classified as Early: Type 1. Finally, if a tract was vulnerable and had experienced no demographic change and only moderate housing market change or none at all, but it lay adjacent to a tract with either high real estate values or high recent appreciation or both, then it was classified as Susceptible. In such a tract, gentrification is likely imminent (assuming that the city’s current economic boom continues), or already underway but not yet showing up in official data because of the time that has elapsed since the data was collected. This classification scheme follows the Bates method precisely. The criteria for inclusion in the five gentrification stages are summarized in the table below.
Adapted from Lisa Bates, “Gentrification and displacement study: Implementing an equitable inclusive development strategy in the context of gentrification, 2013, Table 1, page 31, at https://www.portlandoregon.gov/bps/article/454027.
If a census tract was identified as not vulnerable, it was not classified as a gentrifying neighborhood. It is important to note that simply because a tract is classified as not vulnerable does not imply that it lacks vulnerable people. Rather, such a tract has a lower share of vulnerable people than average. Residential displacement can and does still occur within such areas. One further subcategory recognizes these dynamics: tracts are classified as Continued Loss tracts if they (1) have experienced an above average increase in white and college-educated people from 2000 to 2016, and (2) have housing market values that increased substantially from 1990 to 2016 and are now high. These can be thought of as tracts that have passed beyond the final (Late) stage of gentrification, but that still retain remnant vulnerable populations, many of whose members likely continue to be vulnerable to displacement.
Using census data to make comparisons
To assess whether a given census tract had experienced above average vulnerability, demographic change, or housing market change, we compared it against a wider area. In the case of vulnerability and demographic change, we compared various indicators (five for vulnerability, and two for demographic change, detailed below) against the average for the entire five-county Austin-Round Rock Metropolitan Statistical Area (MSA), which consists of Bastrop, Caldwell, Hays, Travis, and Williamson counties. The City of Austin accounts for just under half of the population of this metropolitan area.
Making comparisons to the MSA is a departure from the original Bates method, which only compared census tracts in Portland to city-wide data. Our procedure intended to capture the metropolitan character of neighborhood change, which involves various populations moving to— or being displaced into—a wide variety of different locations, both inside the City of Austin and outside, within the regional job market.
In another departure from the Bates method, we used a statistical measure called Z-scores to quantify the extent to which a given indicator was above or below the MSA average. By contrast, the Bates method used thresholds: a given indicator was assumed to be above average, or not, based on whether it was above or below a certain level. Z-scores, by contrast, take into account not just whether a given indicator is above or below average, but how much it lies above or below the average.
Census data measured at the tract level is gathered over five-year intervals as part of the American Community Survey (ACS). The most recent tract-level data available at the time we conducted anything other than “non-Hispanic the analysis in this report was for the years 2012 to 2016. By contrast, earlier tract-level data is available from the 1990 and 2000 decennial censuses.
Vulnerability
To assess the vulnerability of neighborhoods to gentrification, we used five variables for measuring the socio-demographics of a given tract as of 2016 (using 2012-16 ACS data):
The first four vulnerability factors are used in the original Bates method. We added the fifth— children in poverty—in response to input from Austin city council members and staff. We considered, but did not include, a sixth indicator: the percentage of residents over the age of 65. As discussed in the prior section, research has found that being an elderly person is not a consistent predictor of vulnerability, if not used alongside other markers of vulnerability (renter, low income, etc.).
Tracts were designated as vulnerable if the Z-score for at least three out of the five vulnerability factors exceeded +0.5. For mapping purposes, we further categorized vulnerable tracts into three subcategories, based on the average Z-scores for all five vulnerability factors: Vulnerable (average Z score was less than +1.0), More Vulnerable (between +1.0 and +1.5), and Most Vulnerable (more than +1.5).
Demographic change
We used four variables to assess demographic change over time between the years 2000 and 2016 (using 2012-16 ACS data). Specifically, we looked at whether there was an increase in the share of residents meeting one or more of three demographic factors: homeowners, higher education, and white. We also looked at changes in median income in each tract.
A tract was deemed to have experienced demographic change if at least two of the four demographic variables had Z-scores that exceeded +0.5, and if the average Z-score for the four factors exceeded +0.5.
Housing market change
Following the Bates methodology, we used three variables to classify tracts on the basis of housing market change. All of them involve median home values reported at the tract level. Note that home value data from the Census and from the ACS is self-reported by respondents and only applies to owner-occupied housing.
Unlike what we did with vulnerability and housing market change, for the housing market analysis we did not compare tracts against the MSA-wide average using Z-scores. Instead, we sorted the 200 tracts within Austin and grouped them into quintiles, i.e., categorized them into five “buckets:” lowest fifth, second lowest fifth, middle fifth, second highest fifth, and top fifth. Because the bulk of recent dramatic home value increases have occurred within the City of Austin, extending this analysis to the entire MSA would have dampened the variation among tracts.
Note that a small number of tracts lack reported median home value data, because they have so few owner-occupied units that the Census cannot release statistically valid estimates for them. In such cases, we benchmarked median rents, rather than median home prices, against the rest of the tracts using quintiles in the same manner.
The three variables we used to classify tracts on the basis of home values were as follows:
- Present home value: Median home value (ACS 2012-2016 data).
- Home value change since 2000: Percent change in median home value from 2000 to 2016 (using 2012-16 ACS data).
- Home value change since 1990: Percent change in median home value from 1990 to 2016 (using 2012-16 ACS data).
Following Bates, we used these variables to identify three types of tracts with notable housing market dynamics:
Intuitively, accelerating tracts are places where the housing market has picked up steam since 2000; appreciated tracts are where this process has already occurred; and adjacent tracts are where this process seems likely to happen soon. Referring back to the gentrification typology discussed earlier in this section, Susceptible tracts have not experienced demographic change, and are in areas adjacent to ones showing signs of housing market appreciation. Early: Type 1 tracts have not yet experienced demographic change but are experiencing an accelerating market. Early: Type 2 tracts are the other way around: they have experienced demographic change but are not yet accelerating and instead are next to an accelerating or appreciating tract. Dynamic tracts have experienced demographic change and are experiencing accelerating market conditions, whereas Late tracts have also experienced demographic change but are in an appreciated housing market state. Finally, among non-vulnerable tracts, Continued Loss tracts, in addition to having recently experienced an increase in their white and college educated populations, are in an appreciated market condition.