Singificant main effect, though no significant difference in pairwise comparisons (adjusted for Bonferroni)  I'm hoping that someone can help me with this. If there is anything I haven't included please let me know!
I am running a one way repeated measures ANOVA (using SPSS), 22 subjects complete testing on 4 occassions to collect a time to exhaustion (TTE). 
TTE data does not violate sphericity (p = .581), so I interpret the trial number as having a main effect on TTE, F(3,63) = 2.883, p = 0.043
main effect SPSS output here
Therefore I would like to investigate what is causing these significant differences. When I investigate the pairwise comparisons (with Bonferroni) none of the trials are significantly different. pairwise comparisons SPSS output here
Therefore my question is: Is it possible for a main effect to be present even though there are no significant differences between any of the trials? 
Thank you!
 A: It is always difficult to diagnose this kind of difficulty without
knowing more about how the data were collected and seeing some diagnostics
of the design. However, I have seen the kind of difficulty you
are describing before, and I will venture some $speculative$ comments on what
the difficulty might be.
Your main effect Occasion is barely significant at the 5% level with P-value 4.3%. This is (relatively weak) evidence that some two of the five occasions may have different population means.
However, I suppose the Bonferroni method is testing each of the ${5 \choose 2} = 10$ pairs at the 0.5% level to ensure (conservatively) no more than
a 10(.5%) = 5% family error rate. The Bonferroni method is conservative because it is based on an inequality.
Because this appears to be a balanced design, perhaps you could use Tukey's
HSD method of multiple comparisons with at 5% family error rate.
Even then there is a small chance you might not identify any pairs of
levels as significantly different. The theoretical possibility of such a discrepancy is that Tukey's method uses a somewhat
different criterion than ANOVA for judging significance. (The closely related Student-Newman-Keuls ('SNK') method of multiple comparisons is
another possibility. It is a bit more aggressive in declaring
significant differences than Tukey's HSD.)
Also, I'm wondering if the same subjects participated on each of
the five Occasions. If so, you should use a design that
recognizes 'Subject' as a random factor (in addition to Occasion).
If the subjects are the same, it is possible that different
Occasions might not be independent, as required by the one-factor
ANOVA. Without seeing the data and understanding more about the
design, I could not say whether this issue is related to your
failure to find significantly different pairs of Occasions
using the Bonferroni method. However, if there is a significant
Subject effect lurking in the background, the true P-value for
Occasion might be smaller, and pairwise comparisons might rise
to significance.
$Note:$ Here is a fake example that may be somewhat like yours. I just ran a one-factor ANOVA with five levels and 10 replications per level with simulated data. Normal population
means are 100, 104, 108, 112, 116 with common population SD 15.
The summary statistics and ANOVA table are as follows:
 Level   N    Mean  StDev  
   1    10   97.80  16.65  
   2    10  109.10  16.55  
   3    10  105.50  22.20  
   4    10  121.90  14.97  
   5    10  113.40  15.62  

 Source  DF     SS   MS     F      P
 Factor   4   3220  805  2.66  0.045
 Error   45  13608  302
 Total   49  16828

Tukey's HSD $just\; barely$ distinguishes between Levels 1 and 4 (not
the two levels with the greatest difference in known population means, as it happens),
and the Bonferroni method finds no paired differences.
Finding a convincing pattern of significant differences among
five sample means when the P-value is barely below 5% can be
a frustrating endeavor. There is no more 'information' than
the data supply, and no amount of fussing with different
multiple comparison methods adds to that available information.
