Skip to first unread message
loic
Sep 18, 2019, 5:23:19 AM9/18/19
to estimationstats
Good afternoon,
I'm sorry to bother you but I have two quick questions about your online app. I think there are small bugs but it could also be because of my small sampling..
Using the Hedges' g method (also with Cohen D) I had the error 'list index out of range' with the data set below. Could it be because the SD for treatment is much larger than in the control that it display this error? Or is there some kind of bug?
Thank you in advance for your help and this nice tool!
Best, Loïc
control:
1.45246
1.19208
1.61360
1.12898
1.27610
Treatment:
1.88
2.33249
0.80159
1.44444
I'm sorry to bother you but I have two quick questions about your online app. I think there are small bugs but it could also be because of my small sampling..
I'm working with your online app to calculate effect size and IC and I
tested different methods.
In some cases, the app does not provide an
answer when I'm performing a Cliff's delta between two groups (it tries to
calculate but no output). It seems to me that it happens when the two
populations do not overlap. My understanding of this method is that the
answer in such case should be -1 or 1. So I am not sure if the app was
bugging or if it was not coded to provide an answer when the two
populations do not overlap.
Using the Hedges' g method (also with Cohen D) I had the error 'list index out of range' with the data set below. Could it be because the SD for treatment is much larger than in the control that it display this error? Or is there some kind of bug?
Thank you in advance for your help and this nice tool!
Best, Loïc
control:
1.45246
1.19208
1.61360
1.12898
1.27610
Treatment:
1.88
2.33249
0.80159
1.44444
Joses Ho
Oct 1, 2019, 2:35:27 AM10/1/19
to estimationstats
Hi Loic,
This looks like a bug, yes. I will let you know when this is fixed!
Joses
Joses Ho
Oct 1, 2019, 6:03:16 AM10/1/19
to estimationstats
Hi Loic,
After closer inspection, your original intuition was right.
We generate bootstraps by sampling with replacement. Because both the control and treatment samples are small (N=5 and N=4 respectively), there will be a bootstrap where the control values are all the same, and the treatment values are all the same. As a results, the standard deviations of both groups are zero.
Because Cohen's d and Hedges g both divide the mean difference by the pooled standard deviation of the groups, (see this link), such a bootstrap sample will produce an undefined result. This jams up the plotting.
I will issue a patch to remedy this. Again, will ping when this is fixed.
Joses
Joses Ho
Oct 8, 2019, 4:37:58 AM10/8/19
to estimationstats
Hi,
This bug has been resolved. Let me know if you spot anything else!
Joses
On Wednesday, September 18, 2019 at 5:23:19 PM UTC+8, loic wrote:
Reply all
Reply to author
Forward
0 new messages