Susanne Widmar pointed out an "estimation" error in a command:
As you can see the missing values are included in the calculation for the 90th percentile. It's obviously a bug. However you may not detect the bug if you're interested in the interquartile range:
I quickly ran the same analysis with the most recent build available on the test page which does not show the problems. . version Current version: 1.1 Release 8 (Build 68) Latest public release 1.1 Release 1 (Build 62) . stattables sex /idr=km SEX N NKM P10KM P90KM Kvinde 490 423 18.00 265.00 Mand 3537 3197 0.00 265.00 . stattables sex /iqr=km SEX N NKM P25KM P75KM Kvinde 490 423 42.00 237.00 Mand 3537 3197 36.00 192.00 . describe km if sex = 1 Variable N=490 Sum Mean (95% cfi) Min p5 p10 p25 Median p75 p90 p95 Max KM 423 50288.0 118.88 109.42 128.35 0.0 0.0 18.00 42.00 73.00 237.00 265.00 277.00 286.00 . describe km if sex = 2 Variable N=3537 Sum Mean (95% cfi) Min p5 p10 p25 Median p75 p90 p95 Max KM 3197 330796.0 103.47 100.23 106.71 0.0 0.0 0.0 36.00 71.00 192.00 265.00 265.00 286.00
Apparently the bug shown has been corrected in build 68. - To avoid such errors a comprehensive test suite has been developed for analysis. Currently the test involves testing of around 650 different conditions and a test for the problem above has been added.
I will update to a later build very soon to remedy the problem.
Known other problems are: Estimation of OR in sparse tables The aggregate (stattables) command can create variable names longer than 10 characters
Apart from that I am not aware of current problems of estimation. If other definitive bugs are known (not requests for new functions) please report to the new bug database, found at Http://www.epidata.dk/php/mantis
Currently development is focused on "long term plans" including rewriting the software for further documentation and testing, experimentation with collaborative editing of documentation and other aspects of the development plan for 2006-2010.
The "in-house" build of analysis has come to 85 now, where build 68 is the latest for public testing.
When the current aspects are completely finalised a new build will be placed for testing.
One of the many new aspects are more user control over the interface, a completely rewritten table command, which includes exact testing of stratified tables, calculation of gamma coefficient for ordinal data and extensive sorting. Working with summarised data is next in line.
Other new aspects are further documentation tools within analysis (value labels, missing value assignment), renaming variables and enhanced control over graphics, e.g. axis and ranges.
Jens Lauritsen EpiData Association