Hi there,

I don't know if it is too late, but I write it in case somebody has the same problem. I was recently experiencing the same issue.

I had two independent variables and therefore I ran a two-way ANOVA for them. After that I ran a second analysis, one-way ANOVA for each of the variables. What I consider important here is to first, with all your knowledge, figure out if it is better to run it separately or as a factorial ANOVA. This can be decided when you think if there could be a relationship between the two independent factors, a relationship affecting the results.

If you are not sure, then I would suggest you use the strategy I used, which is running both analysis and see the error explained by each model. My model of two-way ANOVA evidently explained twice as much error as my model of one-way ANOVA for each variable, therefore suggesting an interaction effect that should be relevant for the results. Thus I chose to use a factorial ANOVA analysis.

Most of the time can be like this, since running only a two-way ANOVA instead of several one-way ANOVA will most likely reduce the family-wise error rate (type I error). But when the independent factors are not clearly correlated, you may want to think about running different one-way ANOVAs.

Hope this helps.

Arnau

Running three one way ANOVAs in this situation instead of a combined factorial ANOVA seems to present the same kind of problem as running multiple t-tests instead of a one way ANOVA, i.e. an increased risk of type 1 error.

Did you include the interaction terms in the 3 factor ANOVA model or not? This is what you should be doing.

Model: y = A B C A*B A*C B*C A*B*C

If you don;t have enough trials to allow for the full factorial model, then you can use a fractional factorial experiemnt.

If you simply used the model: y= A B C, then you need to justify why no interaction terms were included in the model. Let's say this is explained, then finding only one factor significant is more trustworthy than what you can find from running 3 separate one factor ANOVA tests.

I think spss is doing that itself? I just plug in the variables and spss compares different categories...

thats why interaction effect was tested and did not result in a significant value

You need to check which model was used. The ANOVA table will show you if the full model was used or not. SPSS still needs the user to define the model.

As I said. Interaction effects were tested.

I would not use the 3 separate ANOVA tests. Use the 3 factor ANOVA results instead. The results can differ here because the Mean Square Error in the 3 factor model will be most likely smaller than the individual MSE values in the 3 separate ANOVA cases. You can try removing the interaction terms and running the 3 factor model again with y= A B C .

1) Reasoning only in terms of p-value is totally meaningless, you should interpret your statistical results regardless the analysis that produced them

2) It is quite straightforward that running three analyses on three groups separately or running a single analysis that takes into account grouping is quite different (and which one you may better retain depends on the critical interpretation of what you're doing)

I would to suggest would not use the 3 separate ANOVA tests. Use the 3 factor ANOVA results instead. ANOVA seems to present an increased risk of type 2 error.

My best regards!

Hi there,

I don't know if it is too late, but I write it in case somebody has the same problem. I was recently experiencing the same issue.

I had two independent variables and therefore I ran a two-way ANOVA for them. After that I ran a second analysis, one-way ANOVA for each of the variables. What I consider important here is to first, with all your knowledge, figure out if it is better to run it separately or as a factorial ANOVA. This can be decided when you think if there could be a relationship between the two independent factors, a relationship affecting the results.

If you are not sure, then I would suggest you use the strategy I used, which is running both analysis and see the error explained by each model. My model of two-way ANOVA evidently explained twice as much error as my model of one-way ANOVA for each variable, therefore suggesting an interaction effect that should be relevant for the results. Thus I chose to use a factorial ANOVA analysis.

Most of the time can be like this, since running only a two-way ANOVA instead of several one-way ANOVA will most likely reduce the family-wise error rate (type I error). But when the independent factors are not clearly correlated, you may want to think about running different one-way ANOVAs.

Hope this helps.

Arnau