Mikkabel
☆

Belgium,
2021-03-04 17:20
(47 d 03:50 ago)

Posting: # 22242
Views: 466

## Degrees of freedom [General Sta­tis­tics]

Hello everyone,

Recently, we performed a 4-treatment (T1, T2, T3 and T4), 4-period, cross-over study pharmacokinetic study.

Once the statistical analysis performed, I observed that for the comparison between T1 and T3 and between T2 and T4, the number of degrees of freedom for the fixed effect "period" is 2 instead of 3 (which is the case for the other comparisons). Furthermore, "Not estimable" is mentioned in the output of Winonlin.

For information, the statistical analysis was performed using ANOVA model (or equivalent parametric method) with fixed effects usually sequence, subject within sequence, period and formulation.

To be honest, I don't know how to fix it and I don't know how to interpret this.
Could you please inform me if it is an issue related to the software (Winonlin) or is it related to the model used?
Additionally, could you explain me how to interpret these results and what should be the impact on the results obtained form the analysis?

Thanks for your help
Helmut
★★★

Vienna, Austria,
2021-03-04 18:34
(47 d 02:36 ago)

@ Mikkabel
Posting: # 22244
Views: 379

## DF in higher-order Xover (again)

Hi Mikkabel,

» Once the statistical analysis performed, I observed that for the comparison between T1 and T3 and between T2 and T4, the number of degrees of freedom for the fixed effect "period" is 2 instead of 3 (which is the case for the other comparisons).

Not unexpected. I suggest to use the ‘Two-at-a-Time’ approach (see this post, the vignette of PowerTOST, and this article).

» Furthermore, "Not estimable" is mentioned in the output of Winonlin.

As usual. In SAS you’ll get a period or zero (not sure).

» For information, the statistical analysis was performed using ANOVA model (or equivalent parametric method) with fixed effects usually sequence, subject within sequence, period and formulation.

OK.

» To be honest, I don't know how to fix it…

You can’t fix it, sorry.

» … and I don't know how to interpret this.

Good question, next question.

» Could you please inform me if it is an issue related to the software (Winonlin) or is it related to the model used?

The latter. AFAIK, you will see it in any software.

» Additionally, could you explain me how to interpret these results and what should be the impact on the results obtained form the analysis?

Use them as they are.
If you have a lot of stamina see a lengthy thread starting here and another one there.

Dif-tor heh smusma 🖖
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
ElMaestro
★★★

Denmark,
2021-03-04 22:23
(46 d 22:47 ago)

@ Mikkabel
Posting: # 22245
Views: 363

## Degrees of freedom

Hi Mikkabel,

» Could you please inform me if it is an issue related to the software (Winonlin) or is it related to the model used?

It is the design combined with the model, but it is not "a problem" and your data will allow an interpretation. Yes, certain things may be suboptimal and there's always an opportunity to balk at bias one way or another, but the data will surely allow analysis and interpretation.

» Additionally, could you explain me how to interpret these results and what should be the impact on the results obtained form the analysis?

You did not provide info on sequences, dropouts, missing periods etc. You can send me your data, specify the purpose and I will be happy to propose some degree of interpretation.

Pass or fail!
ElMaestro