Just calculate the "distance" of each set of answers from the average,
throw away those that are beyond some limit, recalculate the average.

Play a little with that algorithm and the "some limit" will probably be obvious.



On 6/1/05, Thane Sherrington <[EMAIL PROTECTED]> wrote:
> Here's a a problem I'm wrestling with.  I have a company doing on-line
> performance reviews.  Each employee rates a set of other employees on a
> survey which has six categories with between 3 and 7 questions in each
> category.
> 
> The problem is that there are a couple bad apples who blow through the
> surveys rating someone either all 1s or all 5s, throwing off that person's
> ratings and effectively ruining the value of the performance review.
> 
> My first attempt to stop this was to time the surveys.  People who finished
> them in less than five minutes (the cheaters generally take two minutes)
> got a message telling them to go back and think about their answers and try
> again.  That didn't work because it turned out that several non-cheaters
> print out the review and do it on paper, and then login to enter the
> answers - since they were working from paper, they finished the review in
> under five minutes.
> 
> Then I tried checking each category - if all the answers in a specific
> category were the same, I rejected the review and told them to do it
> again.  No soap - occasionally there are legitimate reviews where one
> category has all the same answers.
> 
> So then I switched to checking the entire survey.  If all the answers are
> same, the survey gets rejected.  It took the cheaters slightly under a
> quarter of a second to figure that one out, as you can imagine.
> 
> The surveys are all anonymous, so I can't simply go to the person entering
> the survey and tell him/her to stop cheating.
> 
> Can anyone think of a way to monitor and block the cheaters?
> 
> T
> 
> 


-- 
 

G. Waleed Kavalec
-------------------------------
Copyright:  G. Waleed Kavalec 2005
This message may be resent and/or republished 
provided the content and this notice are kept intact.

Reply via email to