Standard Deviation

[cdj01]

I am having a problem using the StDev function in Crystal 8.0. I am trying to determing the mean and stdev of lab values for specimens. The mean comes out fine, but the stdev comes out far above what is expected. However, when I run the report for a smaller number of samples, it comes out correctly. And, if I run the report several times, using smaller groups for all the specimens, it comes out correct. Does crystal have a limit to the number of records it can calculate the stdev on? The number of specimen records being used is close to 100,000.
Thanks

 

[cdj01]

To further explain my problem, I just ran a report calculating the mean and standard deviation of test results for about 120,000 specimens. When I ran the report for all the specimens at once, I got a mean of ~14.5 and a standard deviation of ~40. I reran the report using the specimens one month at a time, and came out with means of ~14 and the standard deviation dropped to 9-10. Does crystal have a limit to the number of records it can handle?
Thanks

 

[synapsevampire]

I'm not sure, but here's their whitepaper on SD in CR 8:

http://support.crystaldecisions.com/library/kbase/articles/c2007430.asp

You might try rolling your own, but first try applying the hotfixes.

Good luck with it.

-k

 

[cdj01]

This keeps getting better. I added a hotfix, cr85mainwin_en.zip, which came up when I searched for updates for Crystal 8, hoping this would resolve any issues, and now when I load crystal, it is asking for a keycode again, which would be fine. However it doesnt match the length of the keycode printed on my cd. The field length is totally different. How do I uninstall this?

On another note, I tried calculating the SD by hand using the formula they provided on the help page you gave me above, and here are my results:
n=number of samples=116291
Ex=sum total of samples=1,687,388.60
m=mean value=14.51

sd = square root [(Ex*Ex - n*m*m) / n]
so:
sqrt[(1,687,388.6 * 1,167,388.6 - 116291* 14.51 * 14.51) /116291]

sqrt[(2,847,280,287,410 - 24,483,918.77)/116291]

sqrt[2,847,255,803,491/116291]

sqrt[24,483,887.86]

~4948

I know my math is rusty, but that number is just not possible.