At 6:07 PM -0600 1/17/01, Mike Martin wrote: > >We surveyed a sample of 56 out of over 500 exhibitors, asking booth >attendants to rate Streaming Media West 2000 in terms of 1) Show >Attendance by Non-Exhibitors; 2) Presence of CEO and Decision Maker >Types; and 3) Overall Satisfaction with Dollars Spent (The Value >Question). > >We used a scale of 1-5, five being the most, the best, the grandest >and one being the loneliest number, the dismal, one. We permitted >decimals and fractions for the terminally wavering. > >RATINGS [Rating (# of Exhibitors giving it)]: > >1) Show Attendance by Non-Exhibitors: 1-2 (4); 2.5-3 (20); 3.5-4 (22); >4.5-5 (10) >----Techie Average of 3.11607142857 [for biz dev & marketing-types: >3.1] > >2) Presence of CEO and Decision-Maker Types: 1-2 (3); 2.5-3 (18); >3.5-4 (22); 4.5-5 (13) >----Techie Average of 3.1875 [for biz dev & marketing-types: 3.2] > >3) Overall Satisfaction with Dollars Spent (The Value Question): 1-2 >(2); 2.5-3 (16); 3.5-4 (26); 4.5-5 (12) >----Techie Average of 3.24553571429 [for biz dev & marketing-types: >3.25] > >Interestingly, these ratings nearly follow a bell curve, and are >surprisingly consistent across categories--the 35 attendees we >surveyed provided an average rating of 3.25 as well. Since 2.5 is >true average, Streaming Media West was definitely judged "above >average" by both exhibitors and attendees. This is completely statistically illiterate. Firstly, the average of a 1-5 rating is 1+2+3+4+5/5 = 3, NOT 3.5. For it to be 2.5 you'd need to use a zero to five scale. Your sample is 35 of 500, so expressing averages to 11 decimal places is not techie, it's foolish. Assuming a normal ditribution, the standard estimate of error is 1/sqrt(n) = about 17%. So your mean is 3.25 +/- 0.5. (You could calculate a proper standard deviation from your data and quote that; the point is that the difference between 3.25 and 3 is well within the margin of error) Your bins are not the same size (1-2 has 3 datapoints on, the others have 2.) In any case, binning on an arbitrary scale is not really helpful. You'd be better off giving quintiles (rank the scores in order, break them into even-sized numbers of users, and report the means of those). ============SPONSORED BY REALNETWORKS================ New RealSystem iQ - Bringing Intelligence to Internet Media Delivery Featuring the breakthrough quality of RealAudio 8 & RealVideo 8 ====> To learn more please visit: http://www.realnetworks.com/iq --- To register for Streaming Media Asia/Berlin/West/Europe/East 2001: http://www.streamingmedia.com/upcoming.asp --- You are currently subscribed to skunkworks To unsubscribe send a blank email to [EMAIL PROTECTED] To switch to digest mode, send email to: [EMAIL PROTECTED]
