How will success be judged after a new initiative is taken to increase attendance at LUG meetings?
Success cannot be accurately judged without a dependable record of past attendance. Fortunately, Ted Roche has been recording attendance figures for some time now. How can they be used? The first job is to establish the statistical baseline. We assume (for mathematical convenience) that the attendance has a "systematic" or algorithmicly predictable component plus a Gaussian (Normal) random component added thereto. The systematic portion is beyond this analysis, so assume it is small compared to the average of the random part. If Ted's figures are entered into a calculator or spreadsheet, we can get a statistically reliable (although not necessarily meaningful) estimate of the average attendance and its standard deviation. General variability: Grouping all meeting locations, all seasons of the year, and all presentation topics into one statistic should be done for reference, but it will not give much insight. Variability by location: Assuming Ted's data is conveniently in a spreadsheet, calculate the statistics by meeting location. A low variability might mean the same core of people always attend. On that basis, look at the locations with the highest percentage variability. What are they doing right? Variability by topic: Sorting topics into bins is pretty difficult, and the bins must be meaningful with regard to the goal. Also, with so few data points, the number of bins must be small. One 3-bin starter might be communication (including SSH, protocols, wireless), troubleshooting, and user applications -including "other". Variability by season: One final measure that comes to mind is to statistically process attendance by month of year. Plot the monthly average against month of the year for a quick visual clue to trends. Is there a seasonal variation that must be taken into account before concluding that a given meeting attendance is better due to the advertising or other promotion activity? Unfortunately, there may not be enough data to get meaningful monthly average. Statistic do not become very reliable until there are at least 7 samples of the random variable. That would be 7 years of data. However, the "confidence limits" for the mean and standard deviation can be calculated from as few as 3 points (see Students' T method, Wikipedia). Taking the confidence limits into account, there might be some insight to be gleaned. Conclusions: It is very likely that the standard deviation of these measures is going to be large compared to the average. In such cases, intuitive detection of a trend in almost worthless. Statistics are necessary to see the trend through the noise. How many successive meeting with statistically significant attendance increases are necessary to conclude that an experiment is successful? Despite the slight additional complexity, statistic will be necessary to judge success or failure of the advertising experiments. It would be the correct methodology if gnhLUG is to be run more as a business. However, if gnhLUG is just for fun, as it seems to have been in the past, statistics be damned. Draw any conclusion from the advertising efforts that pleases. Jim Kuzdrall _______________________________________________ gnhlug-org mailing list gnhlug-org@mail.gnhlug.org http://mail.gnhlug.org/mailman/listinfo/gnhlug-org/