Bonferroni-corrected confidence-intervals

[Lim Yonghao]

Dear Listers,
 
I conducted repeated measure ANOVA to access changes across 2 timepoints for 14 dependent variables. Bonferroni adjustments were used to correct for p-values. However, how do we correct for the confidence intervals for estimated mean differences?
 
There is an option to choose what type of adjustment to use (LSD versus Bonferroni) but when i ran the analyses using the 2 adjustments, there wasn't any difference in the CI. Am i using the option correctly?
 
I am using SPSS version 15 for Windows.
 
Cheers,
Yonghao

---

See all the ways you can stay connected to friends and family

[Arthur Kramer]

The Bonferroni procedure is helping prevent you from saying there was a difference between your means when there was no difference—which is what p value of the estimate of the t or F statistic allows you to do. The confidence interval gives you an estimate of where the mean difference may be situated between the populations of interest—which is a parameter. You are not making the same kind of inference with the confidence interval as you are with the test statistic. If you want to be more conservative, make the confidence interval wider.

Arthur Krame

 

"...believe half of what you see and none of what you hear."

                  N.Whitfield and B.Strong

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Lim Yonghao
Sent: Tuesday, March 31, 2009 11:49 PM
To: [hidden email]
Subject: Bonferroni-corrected confidence-intervals

 

Dear Listers,
 
I conducted repeated measure ANOVA to access changes across 2 timepoints for 14 dependent variables. Bonferroni adjustments were used to correct for p-values. However, how do we correct for the confidence intervals for estimated mean differences?
 
There is an option to choose what type of adjustment to use (LSD versus Bonferroni) but when i ran the analyses using the 2 adjustments, there wasn't any difference in the CI. Am i using the option correctly?
 
I am using SPSS version 15 for Windows.
 
Cheers,
Yonghao

---

See all the ways you can stay connected to friends and family

[Lim Yonghao]

Dear Arthur,
 
Thank you for your reply.
 
My next question is: What if the F statistics is non-significant after the correction but the 95% CI of the mean differences does not include 0? This could happen if the p value is around 0.05 (e.g. 0.02). I take your point on increasing the CI but it seems funny if i present 99.996% CI for the mean differences.
 
To put this into context, i wanted to present the results graphically using the estimated mean differences with CI and would like to include information on the significance and width of the CI at the same time. If the ends of the CI touches the 0 mark in the graph, it would correspond to a non-significance depending on the CI i decided to put in. Please correct me if i understood it wrongly (i took this from Geoff Cumming's paper on Inference by Eye in Teaching Statistics).
 
Can we use the critical value of the adjusted alpha but still present it as a 95% CI?
 
Pardon me if i am unclear. I am getting a little confused myself.
 
Cheers,
Yonghao
 

---

Date: Wed, 1 Apr 2009 09:11:19 -0400
From: [hidden email]
Subject: Re: Bonferroni-corrected confidence-intervals
To: [hidden email]

The Bonferroni procedure is helping prevent you from saying there was a difference between your means when there was no difference—which is what p value of the estimate of the t or F statistic allows you to do. The confidence interval gives you an estimate of where the mean difference may be situated between the populations of interest—which is a parameter. You are not making the same kind of inference with the confidence interval as you are with the test statistic. If you want to be more conservative, make the confidence interval wider.

Arthur Krame

 

"...believe half of what you see and none of what you hear."

                  N.Whitfield and B.Strong

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Lim Yonghao
Sent: Tuesday, March 31, 2009 11:49 PM
To: [hidden email]
Subject: Bonferroni-corrected confidence-intervals

 

Dear Listers,
 
I conducted repeated measure ANOVA to access changes across 2 timepoints for 14 dependent variables. Bonferroni adjustments were used to correct for p-values. However, how do we correct for the confidence intervals for estimated mean differences?
 
There is an option to choose what type of adjustment to use (LSD versus Bonferroni) but when i ran the analyses using the 2 adjustments, there wasn't any difference in the CI. Am i using the option correctly?
 
I am using SPSS version 15 for Windows.
 
Cheers,
Yonghao

---

See all the ways you can stay connected to friends and family

---

check out the rest of the Windows Live™. More than mail–Windows Live™ goes way beyond your inbox. More than messages

[Arthur Kramer]

I’m not familiar with Cunningham’s paper, and maybe I don’t fully understand what you are trying to do because I don’t know your design. But a Bonferroni procedure is a post-hoc analysis of a significant F and is interpreted like a t-test. I may be missing something here. I am assuming your original F obtained significance.

 

When I referred to making your confidence interval wider I didn’t mean making it 99% I meant making it more like 90%.

 

Arthur Kramer, Ph.D.

Director of Institutional Research

New Jersey City University

Phone: [phone]

Fax: [phone]

 

"...believe half of what you see and none of what you hear."

                  N.Whitfield and B.Strong

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Lim Yonghao
Sent: Thursday, April 02, 2009 2:45 AM
To: [hidden email]
Subject: Re: Bonferroni-corrected confidence-intervals

 

Dear Arthur,
 
Thank you for your reply.
 
My next question is: What if the F statistics is non-significant after the correction but the 95% CI of the mean differences does not include 0? This could happen if the p value is around 0.05 (e.g. 0.02). I take your point on increasing the CI but it seems funny if i present 99.996% CI for the mean differences.
 
To put this into context, i wanted to present the results graphically using the estimated mean differences with CI and would like to include information on the significance and width of the CI at the same time. If the ends of the CI touches the 0 mark in the graph, it would correspond to a non-significance depending on the CI i decided to put in. Please correct me if i understood it wrongly (i took this from Geoff Cumming's paper on Inference by Eye in Teaching Statistics).
 
Can we use the critical value of the adjusted alpha but still present it as a 95% CI?
 
Pardon me if i am unclear. I am getting a little confused myself.
 
Cheers,
Yonghao
 

---

Date: Wed, 1 Apr 2009 09:11:19 -0400
From: [hidden email]
Subject: Re: Bonferroni-corrected confidence-intervals
To: [hidden email]

The Bonferroni procedure is helping prevent you from saying there was a difference between your means when there was no difference—which is what p value of the estimate of the t or F statistic allows you to do. The confidence interval gives you an estimate of where the mean difference may be situated between the populations of interest—which is a parameter. You are not making the same kind of inference with the confidence interval as you are with the test statistic. If you want to be more conservative, make the confidence interval wider.

Arthur Krame

 

"...believe half of what you see and none of what you hear."

                  N.Whitfield and B.Strong

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Lim Yonghao
Sent: Tuesday, March 31, 2009 11:49 PM
To: [hidden email]
Subject: Bonferroni-corrected confidence-intervals

 

Dear Listers,
 
I conducted repeated measure ANOVA to access changes across 2 timepoints for 14 dependent variables. Bonferroni adjustments were used to correct for p-values. However, how do we correct for the confidence intervals for estimated mean differences?
 
There is an option to choose what type of adjustment to use (LSD versus Bonferroni) but when i ran the analyses using the 2 adjustments, there wasn't any difference in the CI. Am i using the option correctly?
 
I am using SPSS version 15 for Windows.
 
Cheers,
Yonghao

---

See all the ways you can stay connected to friends and family

 

---

check out the rest of the Windows Live™. More than mail–Windows Live™ goes way beyond your inbox. More than messages