Hello,
I'm analysing data from a protocol which in fact works under the same structure as described in https://www.stata.com/support/faqs/s...nova/#wsanova2, case "One between-subjects factor with two repeated-variables example from the anova manual entry". See image below to understand the case:
I've conducted the repeated-measures ANOVA as described in the webpage.
In the same time, I've conducted a simpler ANOVA just for "period 1", which can be considered independently as subjects in Period 1 are not biased by a previous period, unlike Periods 2 and 3.
Obviously, the factors' significance are not the same between the two ANOVAs. And in particular, "Type of dial" factor (or the equivalent factor in my protocol) is not significant in the simple ANOVA "period 1" (Prob>F = 0,51) but the Types of dial from "Period 1" stemming from the whole repeated-measures ANOVA are significantly different. This can be seen with factors' margins, and especially the interactions "Type of dial"#"Period 1", whose confidence intervals don't overlap, even at 99%.
If considering "Period 1" independently remains relevant, how to deal with the mismatch originated by the comparison with a bigger-scope ANOVA? What summarized and consistent test should I run which would consider the "unidirectional" effect of time, so that Period 1 variance analysis doesn't take into account later Periods (which would have no sense) while other Periods variance analyses do take into account previous Periods' variance?
Thank you very much for your help.
I'm analysing data from a protocol which in fact works under the same structure as described in https://www.stata.com/support/faqs/s...nova/#wsanova2, case "One between-subjects factor with two repeated-variables example from the anova manual entry". See image below to understand the case:
I've conducted the repeated-measures ANOVA as described in the webpage.
In the same time, I've conducted a simpler ANOVA just for "period 1", which can be considered independently as subjects in Period 1 are not biased by a previous period, unlike Periods 2 and 3.
Obviously, the factors' significance are not the same between the two ANOVAs. And in particular, "Type of dial" factor (or the equivalent factor in my protocol) is not significant in the simple ANOVA "period 1" (Prob>F = 0,51) but the Types of dial from "Period 1" stemming from the whole repeated-measures ANOVA are significantly different. This can be seen with factors' margins, and especially the interactions "Type of dial"#"Period 1", whose confidence intervals don't overlap, even at 99%.
If considering "Period 1" independently remains relevant, how to deal with the mismatch originated by the comparison with a bigger-scope ANOVA? What summarized and consistent test should I run which would consider the "unidirectional" effect of time, so that Period 1 variance analysis doesn't take into account later Periods (which would have no sense) while other Periods variance analyses do take into account previous Periods' variance?
Thank you very much for your help.
Attached Files
Comment