Many a diary has slipped off the recent list since I last
updated the description of impact. It's time for another.
To review, impact is the combined measure of recommendations and comments that I have used in daily and weekly lists summarizing recently published diaries. Impact is also tracked and used in the search function for diaries.
Until recently I had only the number of comments available for each diary, not the number of commenters, or as I sometimes call them commentators (thus 'tators ... sorry about that). That's why I used number of comments in the definition of impact - it was the best I had.
Since I'm now in the process of moving the daily lists on site, it is possible to count number of commenters and even to combine recommenders and commenters togther to yield all those who've connected either way with a diary. These new numbers call for consideration of an update to the definition of impact.
The update should result in a scale compatible with the old system. It should provide a better balance between approval (recommendations), overall interest (connections), and intensity (repeated comments by the same commenter). Based on past experience we want to emphasize approval and interest and downplay intensity. One man one vote.
I've considered many different updates to the current impact standard, of which I'll describe three.
- Minor revision. This would entail no change in the method of calculation but would correct the list of diaries in the standard set, Cheers & Jeers diaries posted by Bill In Portland Maine in the first quarter of 2005.
- Replace number of comments with number of commenters. This is the method I've used in recent lists in which an additional value for impact is reported. This method tends to discount intensity of interest, since it discards multiple posts by the same person to one diary.
- Replace number of comments with number of connections. This is a further advance using the number of those who either recommended or commented. This prevents double counting of those who both recommend and comment.
I've decided to go with method 3, in which connections are used rather than comments or commenters. It uses numbers which most clearly demonstrate approval (recommendations) and interest (connections) without over emphasizing individual intensity. Diaries that should have been part of the standard c&j q1 2005 set but were accidentally left out were reinstated.
The new formula for impact is as follows, where cnx will be the abbreviation for the union set of recommenders and commenters and the average for nrec and cnx in the standard set is given to 5 digits.
impact = sqrt( ( (nrec/78.927)^2 ) + ( (cnx/136.38)^2 ) / 2.0)
Several additional methods were considered but discarded and are not further described. The following table shows the statistics of the c&j q1 2005 set with impact as presently calculated (impact) and as calculated by the 3 methods described above (i1,i2,i3).
stat | nrec | ncom | tators | cnx | impact | i1 | i2 | i3 |
n | 55 | 55 | 55 | 55 | 55 | 55 | 55 | 55 |
min | 60 | 264 | 80 | 111 | 0.81 | 0.82 | 0.80 | 0.79 |
max | 98 | 541 | 124 | 165 | 1.24 | 1.24 | 1.23 | 1.23 |
median | 79 | 409 | 101 | 135 | 0.99 | 0.99 | 1.01 | 1.00 |
avg | 78.9 | 412.1 | 101.5 | 136.4 | 1.00 | 1.01 | 1.00 | 1.00 |
stdev | 10.4 | 71.2 | 11 | 14 | 0.11 | 0.11 | 0.11 | 0.11 |
Daily lists will use the newly revised method. Search will continue to report and use the older impact method for a short period until all indices can be revised.