univariate analysis of variance

[barry]

I used univariate analysis of variance to check for differences between age groups on a measure. The F value was .011, which was significant, suggesting there are differences between the means of the groups. However, when I did post hoc, TukeyHSD, there were no significant differences between any of the groups. What does this mean?

[quark]

Hi Barry,

Welcome to Talk Stats. p-value of 0.011 is highly significant, and the Tukey test should show a difference between at least one pair of groups. Have you tried plotting the data by group? Either a boxplot or scatterplot will show you how the groups differ. If you provide more details such as the anova table and Tukey results I'll try to see where the problem is.

[barry]

Hi Barry,

Welcome to Talk Stats. p-value of 0.011 is highly significant, and the Tukey test should show a difference between at least one pair of groups. Have you tried plotting the data by group? Either a boxplot or scatterplot will show you how the groups differ. If you provide more details such as the anova table and Tukey results I'll try to see where the problem is.

Hi quark,
If you tell me how to contact you I will email my SPSS output file for you to consider.
Thanks, Barry.

[quark]

Barry,

My email is webmaster[at]talkstats.com. Thanks.

[quark]

Barry,

I think the results is largely due to the extreme difference in the sample sizes, as listed below.

group sample size
1 3776
2 726
3 365
4 189
5 165
6 83
7 62

Because total sample size is mostly from group 1. The difference between any pair of groups will be "exaggerated" in the anova results, ie it is likely to have significant difference due to the large overall sample size alone.

The p-values between groups 3 6, and 2 6 are below 0.10. I think these are marginally significant differences.

Please let me know if you have further questions.

[quark]

Thanks for your quick and informative reply.
I have done my analysis again, this time not including group1, as you say it is disprortionally large compared with the other groups. So, using groups 2,3,4,5,6,7, I still get a significant F and the Tukeys show no significant difference between any pair of groups.
Could you comment on this please? I am attaching this new printout.


Hi Barry,

Thanks for the new output. The problem with sample size still exisits (to a lesser extent) after removing group 1, since group 2 with n=726 now has a predominant effect on overall sample size.

Although the Tukey tests are not significant, the p-values of groups 3 6 (0.064), and 2 6 (0.073) are smaller than the previous analysis. ie the differences between the pairs are more significant after removing group 1.

I think if you remove both group 1 and 2 then the p-value for 3 6 will be less than 0.05. :)

[penyrallt]

i have used a univariate analysis of variance 2 by 2. i wandered what the error means in the results. i have 199 participants and my error is 195 this seems high to m! F(1,195)=197.37,p<0.05! what does the 195 and 197.37 mean? help much appriciated!

also what further ananysis should i perform on the sig results?