Area Median Income Data

Alex

The CHAS data is s special tabulation of ACS data done for HUD. I do not think that CHAS data exists for single years, only for period of 3 or 5 years. At present it is only offered for using five year data. Since different area median incomes apply through out the country it would be very time consuming to construct a version from the one year PUMS data set. This could be done but but it would be very difficult. You can learn more about the CHAS data here:

www.jstor.org/.../20868636

Cliff Cook
City of Cambridge, MA

That's unfortunate, that would have made our work so easy! But thanks for letting me know, this will still be useful.

I would encourage you not to give up too quickly. As Cliff pointed out, it isn't trivial to do this, but neither is it so difficult that it isn't worth trying (depending on how much you want the data). To demonstrate, I created & uploaded a tabulation for you that dimensionalizes every household in the 2014 PUMS file by HAMFI, using the breakouts that were in the link you provided (<30%, 30-50%, etc.). I used a two step process. First, I dimensionalized the family incomes (fincp) of all occupied households by PUMA (the smallest geographic area available in PUMS) and state (because PUMAs are duplicated across states). Then I used a spreadsheet to calculate the average income for each area (PUMA in this case, but the same could be done for other areas, such as states). Then I used that file to create a new variable (call it average area income) for every household.

In the second step, I created the tabulation that I uploaded, comparing each family income to the average for their PUMA (which I had derived in the first step), using the divisions that were in the link. If the PUMS file will work for you (meaning that PUMAs or geographies that can be made up from PUMAs are acceptable) I see no reason why this can't be done. The hardest part is probably creating the HAMFI, which I've already done if you are to use PUMAs. Things like kitchen and plumbing facilities are pretty trivial to include.

Caveat: After I did this, I realized that the 'M' in HAMFI is supposed to be the median, and I used the mean. I question the value of using the median in this case from the PUMS, because the family income is topcoded (so the mean is artificially set). At any rate, if you are interested in pursuing this & want to use the mean, it can be done. The thing to figure out first is whether or not PUMS geographies will work for you, and will the 2014 file be sufficient (it is the only single-year file that I have available).

I would encourage you not to give up too quickly. As Cliff pointed out, it isn't trivial to do this, but neither is it so difficult that it isn't worth trying (depending on how much you want the data). To demonstrate, I created & uploaded a tabulation for you that dimensionalizes every household in the 2014 PUMS file by HAMFI, using the breakouts that were in the link you provided (<30%, 30-50%, etc.). I used a two step process. First, I dimensionalized the family incomes (fincp) of all occupied households by PUMA (the smallest geographic area available in PUMS) and state (because PUMAs are duplicated across states). Then I used a spreadsheet to calculate the average income for each area (PUMA in this case, but the same could be done for other areas, such as states). Then I used that file to create a new variable (call it average area income) for every household.

In the second step, I created the tabulation that I uploaded, comparing each family income to the average for their PUMA (which I had derived in the first step), using the divisions that were in the link. If the PUMS file will work for you (meaning that PUMAs or geographies that can be made up from PUMAs are acceptable) I see no reason why this can't be done. The hardest part is probably creating the HAMFI, which I've already done if you are to use PUMAs. Things like kitchen and plumbing facilities are pretty trivial to include.

Caveat: After I did this, I realized that the 'M' in HAMFI is supposed to be the median, and I used the mean. I question the value of using the median in this case from the PUMS, because the family income is topcoded (so the mean is artificially set). At any rate, if you are interested in pursuing this & want to use the mean, it can be done. The thing to figure out first is whether or not PUMS geographies will work for you, and will the 2014 file be sufficient (it is the only single-year file that I have available).

John

I think what you outline could be an interesting and worthwhile effort to undertake. However, it would not yield the same result as using the HAMFI assigned to metro areas and non-metro areas by HUD. these figures are devised using a variety of inputs and adjustments, not just from ACS data alone. They vary across the country and probably within each state. (I am only familiar with Massachusetts.)

I am unclear as to why topcoding of the income data would undercut the utility of median calculations. I would think that the median would suffer far less from the effects of topcoding than a calculation of the mean value.

Hi Cliff,

I understood from what I had read earlier in the thread that it could be calculated from ACS - but if there are other inputs required, this approach wouldn't work very well.

I believe you are also correct in your second point, thanks for pointing both of these out.

Please pardon me for being late to this discussion; I've been on the road. A couple of points worth noting:

1. Cliff is correct; the top-coding has no impact on the median. It does, however, affect the mean, especially in high income areas.

2. The data can only really be updated as new ACS data are released. Large counties can be updated annually - but that's only those of 65K+ population and, for practical purposes, more like 100K given the MOE problems. Now that we no longer have 3 year data, smaller counties only have truly new data sets once every 5 years. You could do annual calculations, but since 80% of the data remain the same from year to year, it's hardly worth the bother.