Mantel test Not significant at strata level BUT significant at aggregated level

[dcortell]

I'm running a test where, within a marketing event, for the group of invited people (Attendee VS  Not Attendee), I want to test if the N of new opportunity generated has been significatively different between the people attending and people that were invited but not partecipated (a new opportunity could be generated in both groups)

With the following code:

proc freq data=lead_att (where=(outcome_rev2 is not missing) );

tables inter_group*yy_event*outcome_rev3/cmh chisq nocol nopercent trend measures; /*test the trend */

tables inter_group*yy_event*outcome_rev3/cmh nocol nopercent noprint scores=modridit ; /*test existance corr*/


exact mhchi measures trend; 
run;

I get the results below. The Cochran-Mantel Exact Test is significant at aggregated level, by controlling the 'Attending the event' variable, with a values of 4.8065 (Non Zero Correlation).

However, looking at the individual STRATA, the Exact Mantel test is not significant in any of the two strata levels.

How Should I interpret this result? I would expect that as both the strata are not significant, too the aggregated result should follow the same result.

Any suggestion in the interpretation is really appreciated

[Ksharp]

It is normal result. No surprise for you .

[dcortell]

Hi.

I'm not questioning the result. I'm more questioning how I should interpret it.

It seems to me like , taking into account the individual strata, as the individual Exact Mantel Tests are not significant, then there is not enough evidence to refuse the hypothesis that the share opportunities generated (in %)  are equal  between each event edition.

But then, If we analyze the aggregated result, then the Exact Mantel test suggest us that indeed exists an association between event edition and share of opportunities generated.

The thing I'm missing is how to connect the two results at a theoretical level

[Ksharp]

No connection. You just watch data at two different side.