Non Significant ANOVA results - need to explain why.

#1
I was given data to work on for a lab report. I did a 2 x 2 between subjects ANOVA (if it matters n=187).

Main effect of IV1 was not significant.
Main effect of IV2 was not significant.
Interaction was not significant (although was .62 which I'm supposed to report as "marginally significant").

All prior research suggests that both main effects should have been significant.

What are some reasons that my main effects wouldn't be significant? What are some reasons that my interaction was not significant? What should I recommend be done to obtain a significant result?

I am so lost. I have no idea why these results have happened (the are definitely correct though).

Thank you for any help at all!!!!

Luna
 

bugman

Super Moderator
#2
Without any background information about the experiment, it is basically impossible for any of us to help explain why!
 
#3
I thought I replied to this but it seems to not be showing. I'm going to reply again, if it posts twice sorry.

Thanks for letting me know more information is needed! I wasn't sure if there was a really obvious basic answer that applied to all non significant anova results.

I have included more information below. Essentially all research suggests that there should have been a significant main effect of IV1 and IV2 on the DV (and there was likely to be a significant interaction). Please let me know if more specific information is needed.

Question asked 1. How do the administrator’s comment and eyewitness confidence influence evaluators’ ratings of eyewitness accuracy, independently and in combination? That is, we are examining the main effects and interaction of our independent variables on evaluators’ ratings. In terms of the interaction, we are asking whether the two variables have an additive or interactive effect?

Main effect of eyewitness confidence --> Ratings of eyewitness accuracy?
Main effect of administrator comment --> Ratings of eyewitness accuracy?
Administrator Comment x Eyewitness Confidence --> Ratings of eyewitness accuracy?

It is a 2 x 2 ANOVA because there are two independent variables, each with two levels: confidence (low or moderate) and administrator comment (neutral or confirmatory). Thus, there were four conditions in total:
1. Low eyewitness confidence and neutral administrator comment
2. Low eyewitness confidence and confirmatory administrator comment
3. Moderate eyewitness confidence and neutral administrator comment
4. Moderate eyewitness confidence and confirmatory administrator comment

The design is between groups because each respondent only participated in one condition. We are interested in testing whether the effect of low or moderate eyewitness confidence on ratings of eyewitness accuracy varies according to whether the administrator made a neutral comment or confirmatory comment. This means we need to see if there is a significant interaction between confidence (low vs. moderate) and administrator comment (neutral vs. confirmatory) on eyewitness accuracy ratings (evaluators’ perception that the eyewitness made an accurate identification).

Results

The results above show that the main effect of IV_Conf is not significant. That is, eyewitness confidence did not influence perceptions of eyewitness accuracy.
The results above show that the main effect of IV_AdComm is not significant. That is, administrator comment did not influence perceptions of eyewitness accuracy.
The results above show that the IV_Conf*IV_AdComm interaction effect is marginally significant. That is, there is a marginally significant interaction between eyewitness confidence and administrator comment but the effect did not quite reach the level required for statistical significance.
 
#4
I'm wondering if it's something like "you needed to run a *** test" or "sometimes anova can't be trusted because ***" or "you needed more people and ***"

I really have no idea and it's driving me crazy because I can't find any information about this. I don't know why the results weren't significant, they should be! It's driving me crazy -_-
 

Karabiner

TS Contributor
#5
You should maybe post the 4 group means and stadrad deviations (by the way, were the assumptions of ANOVA met?).

All prior research suggests that both main effects should have been significant.
Prior resear maybe overstated the effects. This is a common problem, because non-significant results don't get published, so the picture could be incomplete. And/or there's something different in your design, compared to former research (e.g. less effective experimental manipulations? a different population from which the sample is drawn, and who react uncommonly?).

Just my 2pence

K.