Calculating Small Area Variation Analysis (SAVA) Statistics
1. My SAVA published papers.
http://works.bepress.com/cgi/viewcontent.cgi?article=1011&context=paula_diehr
http://works.bepress.com/cgi/viewcontent.cgi?article=1010&context=paula_diehr
http://works.bepress.com/cgi/viewcontent.cgi?article=1009&context=paula_diehr
http://works.bepress.com/cgi/viewcontent.cgi?article=1008&context=paula_diehr
http://works.bepress.com/cgi/viewcontent.cgi?article=1007&context=paula_diehr
http://works.bepress.com/cgi/viewcontent.cgi?article=1053&context=paula_diehr
2. My SAVA Technical Reports
http://www.bepress.com/uwbiostat/paper328/ (more about the CVA)
http://www.bepress.com/uwbiostat/paper116/ (MAF)
3. Carlisle paper, variation among L.A. zipcodes
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1070349
4. Spreadsheet, reproducing the calculations for variation in angioplasty among LA zipcodes. If printed, it will show the summary calculations which are in columns A-E rows 292-304. The values from the Carlisle paper are also shown for reference (and should be identical).
http://faculty.washington.edu/pdiehr/SAVA/SAVA_Carlisle_PDiehr.xls
5. Simplified spreadsheet that can be used by others to calculate their own SAVA statistics is here:
http://faculty.washington.edu/pdiehr/SAVA/cva_simplified08.xls
- Save a copy of the spreadsheet under a new name.
1. Enter the multiple admission factor (MAF) in space C279. It is currently set to 1.0, but it should be set to a higher number if persons may be readmitted within the same year. Failure to account for the possibility of multiple admissions per person gives tests and descriptive statistics that are incorrect. You can calculate the real MAF if you have a source of person-level admission data with unique identifiers. That is explained in Cain K et al: Testing the null hypothesis in small area analysis. Health Serv Res 27:267-294, 1992. Estimated values for the MAF for many DRGs are given in the following reference, and an approximate MAF may be used if you have no other source.
http://www.bepress.com/uwbiostat/paper116/ (MAF)
- The spreadsheet is set up to handled up to 20 subgroups, probably one for each age-sex stratum that is desired). The data values are currently the Carlisle values, identical in each subgroup. Enter your data in columns U and following. For example, for subgroup 1, enter the population for area # 1 in U7, and the number of admissions in V7. The values in W7 and X7 are calculated automatically. For subgroup 2, put the data for the first area in Y7 and Z7, and so on. (If you have only 1 group, put it in the subgroup 1 position). (Probably a good idea to save now).
- Erase (do not delete) all of the columns that are not needed through column CW. For example, if you had only 2 subgroups, you would erase everything in columns AA through CW. (Probably a good idea to save now).
- Erase (do not delete) all of the Carlisle data that is remaining. For example, if you have only 20 small areas, you would remove all of the old data for areas 21-236 (starts in t27, continues to t242). Don’t erase lower than row 242.
- To check whether you have done this correctly, look at A245. This is the calculated value of “K”, and it should be the number of small areas that you have.
- If this works, then you may print the output. Your values will not agree with the Carlisle values that are listed there.
I’m not a very accomplished spreadsheet maker. If you make any improvements to this spreadsheet, please let me know. Also, if you publish a paper using this information, please cite the following paper: