MS browser stats based on dud data
Six million, plus or minus two million? Oh dear...
MS on Trial Much of Microsoft's defence has relied on some data - the so-called MDC data set - that is now realised to be seriously duff. Microsoft's economics expert, Dean Richard Schmalensee, admitted that he might not currently be up to speed on statistics, and indeed he wasn't. Schmalensee reworked data that should have been presented with confidence levels at a certain probability, following criticism from the DoJ's economist witness Fisher about this. Schmalensee should not have said that "One reason we work with quarterly data, instead of monthly data, was to reduce margins of error and... to increase sample size", because aggregating data cannot decrease the error from invalid data. In fact the revised exhibit shows very well why Microsoft was reluctant to use confidence levels: in one graph, at the end of the study period the result is 6 million Netscape browsers, plus or minus around 2 million, at 95 per cent probability. This is meaningless, especially when the data is examined: the telephone questionnaire shows the fundamental design errors in the survey. It was disingenuous, and deliberately misleading, for Schmalensee to claim that "I used the MDC database because that's really the only available data that relates to how - that tells us how individuals got the browser." The data are incapable of showing anything useful, and Microsoft knew this. ® Complete Register Trial coverage
Sponsored: RAID: End of an era?