ANOVA REPEATE MEASURES - HOW TO INTERPRET CONTRADICTORY RESULTS BETWEEN ANOVA AND MULTIPLE PAIRWISE COMPARISONS (post hoc)?

[Tom Salvitti]

Good mornig to everybody.

1)HOW TO INTERPRET CONTRADICTORY RESULTS BETWEEN ANOVA AND MULTIPLE PAIRWISE COMPARISONS (post hoc)? I ran a model and i find Non-significant ANOVA with significant multiple pairwise comparisons...i MEAN HOW to write this in an article??? 2) Another question How can I report results of Outcome, timepoint and the interaction between outcome and timepoint ? (see the output below) (table 1)

table 1

Source Partial SS df MS F Prob>F

Model .494415648 19 .026021876 6.40 0.000000

Outcome .063395043 3 .021131681 5.20 0.0016
Timepoint .294629916 4 .073657479 18.11 0.0000
Outcome#timepoint .37227184 12 .031022653 7.63 0.0000

Residual 1.27713207 314 .0040673


thanks to everybody

[Carlo Lazzaro]

Tom:
as per FAQ, your chances of getting helpful replies is conditional on posting what you typed and what Stata gave you back (via CODE deliters, please). Thanks.

[Tom Salvitti]

thanks a lot...but the thing doesn't change...Nobody answer me

have a nice day

[Joseph Coveney]

A couple of observations:

That ANOVA table doesn't look like one for repeated measures.

It's not surprising that pairwise multiple comparisons returns statistically significant test results when the interaction term of the corresponding ANOVA doesn't.

Why is a variable called Outcome in the model as a predictor? Is it short for outcome_type or something? If the latter, you might want to model the residual variances separately by outcome type (using, say, mixed).

A recommendation:

Another question How can I report results of Outcome, timepoint and the interaction between outcome and timepoint ?

help margins
help marginsplot

[Carlo Lazzaro]

Tom:
when nobody reply to me, most of the time it means that my question was ill-posed.
Please, take a loook at the FAQ.
In addition, as Joseph pointed out, reading -outcome- among predictors is admittedly puzzling.

[Marcos Almeida]

Carlo and Joseph already provided a sure guide.

To improve the chances of getting an insightful reply, you should (as per FAQ):

a) inform exactly what you typed;

b\ post code and results within code delimiters or use dataex for that matter;

c) inform the results of the post-hoc tests;

d) describe succintly the model.

That said, there are weird things in this model to solve before "writing this in an article".

To start, you said you've got a non-significant ANOVA, albeit the output highlights big F stats and significant p-values.

The residuals are proportionally large, hence the odds are this model is not fully specified.

Last but not least, there is a predictor named "outcome".

Hazarding a guess, and sorry if I'm wrong, I can only envisage two hypotheses: either a quite misleading term for a predictor was selected, which is unfortunate, or the introductory knowledge of ANOVA is lacking, since it was apparently decided to inlcude the outcome as a predictor.

There are other aspects to comment (such as the use of interaction terms, factor notation, appropriate command to perform a repeated-measures ANOVA, etc.), but I believe this information is enough to guide you through this issue from now on.

Hopefully that helps.