The first is called 'Multivariate Tests'. This gives 4 p-values for each contrast, however as I only run univariate analyses these 4 values are always the same.

The second is called 'Tests of within subjects effects'. This is for where the Sphericity assumption is violated. Again their are 4 different p values for each contrast, but they only vary when Sphericity is violated.

When running a ANOVA I usually use the p value statistics from the multivariate table unless Sphericity is violated. On the ANOVA I've just run Sphericity has been violated for a particular contrast, however I've noticed that the 'Sphericity assumed' p value on the second table is smaller than the multivariate p values (.03 vs .09). Furthermore all the 'sphericity adjusted' p values are also smaller than the multivariate p value.

My questions are 1) why are the p values smaller on the second table, even when sphericity is assumed and 2) which p value should I report? If I use the G-G to adjust for spherecity I get a p of 0.038 which is significant, while if I use the mulitvariate test I get p=0.09 which isn't significant (arguments about the arbitariness of p=0.05 notwithstanding)! I ought to report the significant G-G value, but I'm a bit worried that the multivariate test is actually showing the contrast as non-significant.