Does anyone know of an algorithm to calculate purchasing power on a postcode level? What source data is needed? Purchasing power on admin. level, population?


You need census data, the spatial files for postal code geography (census tacts, dissemination areas ..... some form of spatial parameter) and you should read up on the Cluster Analysis Tool in ArcGIS. The database containing the the census data includes household income, family size, etc, etc (most of the data you will need). The cluster analysis will factor in all relevant variables in the analysis (which you will need to determine!!). I have only done this with a statistical package called SPSS for Windows; never "just" in ESRI. That said, I don't know how accurate this technique would be in the ESRI environment. SPSS is a VERY powerful tool. Statisical analysis is an iterative process; be mindful of this. You need to carefully choose your variables, run the analysis, determine which variables should be removed, and re-run the technique. You are looking for an R-squared value of at least 80% to achieve any sort of accuracy in the analysis. You will need to perform a table join following the analysis to add the output to the attribute table in your map (if you go the SPSS route). That said, you will need to maintain a commin identifier in the output table (ie - census tract ID#).

As far as I know, census data are free in the USA, but cost money in Canada if it's being used "for profit" (unless it's for a school project - then your Geography Department will have access). You will need a solid understanding of Census Geography, and the statistical techniques as well. You might want to read up on Principle Components Analysis as well. This will produce "components" as opposed to "variables", which can augment you analysis a great deal. If you've never done anything like this (stats/quantitative methods) you may find it difficult to grasp without the benefit of formal schooling in this area. I don't know what your background is, and I don't know what kind of a budget you have. What I can tell you, is that purchasing an older version of the software online will save you a lot of money, and a version that's a few years old will do everything you need. Do some reading, and if you have any questions, or need clarity, feel free to ask. Statistic/quantitative methods was one of my strengths in University (I love that kind of math!!). I also did a number of courses in demographics. You're working with some really fun and interesting data!!

Hope this helps -- cheers

  • +1 For the focus on the question (algorithm and data) and the discussion of statistical analysis. Please be cautious, though, in using R-squared as a criterion in any statistical analysis. It reflects the variation in your independent variables as much as anything. Thus, a country-wide analysis could (in principle) yield a much higher R-squared than a regional analysis but have greater error. Also, cluster analyses are usually not appropriate for variable selection. This is a big topic: you can read a lot about it on our sister stats.SE site. – whuber May 23 '11 at 15:17
  • 1
    A couple of open-source stats software packages you might consider using to apply Dano's suggestions are PSPP (gnu.org/software/pspp) and R (r-project.org). – RyanDalton May 23 '11 at 17:30
  • Good point whuber. He will definitely have to be cautious in terms of validity (on the R-squared thread). What about if the Principle Components Analysis (PCA) were done first, then "that" data were run through a Mulitvariate Regression Analysis, then mapped (housing value could be the "dependant variable"). The data would speak for itself, would it not? A highly discriminant Cluster Analysis on the PCA data would augment this study too. The dendrogram would be screaming with indicators and answers, and he could map that too. Thoughts? – Dano May 24 '11 at 14:00
  • 1
    @Dano I confess you have lost me. I'm still stuck at the earlier step of figuring out how one could estimate purchasing power from Census data. There's no such variable in US Census data AFAIK, so you need some theory about how to construct it from related data like annual income and markers of socioeconomic status--or, as others have pointed out, you pay big bucks for a proprietary database from somebody who already has a theory and did the calculations. Then you can start to have fun with PCA, cluster analyses, and the like! – whuber May 24 '11 at 23:08
  • @ whuber - this is all part of the fun if you're a statistics hound. It is an exceptionally itereative process (to say the least). Stages of the "entire" process, inevitably, will need to be revisited, approached will have to be changed ......> – Dano May 25 '11 at 18:05

I am sorry I dont know of any specific algorithm, but I could find some links which might be helpful.

Purchasing Power in Europe

A New Method for Classifying Customer Purchasing Power

Doubly Truncated ARMA-GARCH Model and MCMC Algorithms

Hope it helps...

  • +1 This interesting collection of references covers the gamut from introductory to highly technical and gives a better overview of what purchasing power is and what kinds of data might be needed to estimate it. – whuber May 23 '11 at 15:20

You'd need something like an experian data cd, which are insanely expensive, about demographics, which you'd then have to query into the data you required.

As I said, it is very expensive data. I used to design drive time polygons which would be used to query this data, in order shops could be better located according to who they were after tapping into.

I'd call someone like Pitney Bowes MapInfo who would be able to supply the current data, or at least give you a supplier name.

  • It is 'insanely' expensive due to royal mail (postcode PAF file) Ordnance Survey maintain geographic correctness and Experian added value (purchase power, population, census) all adding their 100% profit (each) to the cost of the data – Mapperz May 23 '11 at 18:08
  • I know, I think back in 2001 they were asking £100k or the one cd – Hairy May 23 '11 at 18:35

Depending on which country you are in there may be census microdata available (e.g. PUMS in US, SARs in UK, Canada has some too I think). You may need to find some finance data too and do a statistical merge to get the variables you need. Then you can look at microsimulation methods to estimate the population at post code level (see http://jhi.sagepub.com/content/12/1/65.short for example).


To construct a purchasing power variable, call it PPP_zip, for USA zipcode areas, the key variables could be as follows:

  • CPI_Metro, Consumer Price Index by Metro area, source BLS
  • Income_USA, Median income USA, source Census
  • Income_zip, Median income by zipcode, source Census
  • Home_Zip, Zipcode home property values
  • Home_Metro, Metro area home property values


PPP_zip = Income_zip / Income_USA * 1/ ( CPI_Metro * Home_Zip / Home_Metro )

Since CPI is probably not reported at the zipcode level, I think finding a good zipcode local cost adjustment for CPI_Metro is the challenge. I am using Home_Zip/Home_Metro above, but there might be better data available to represent the living cost differences between zipcodes of a metro area.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.