Why do the results of the global test contradict the results of the maximum likelihood ratio?

[juvepirlo]

proc surveylogistic data=sasuser.begin nomcar;
strata sdmvstra;
cluster sdmvpsu;
weight WTFOL2YR;
domain category('1');
class serumgroup/param=ref ref=first;
model diabete(desc) = serumgroup;
run;

As shown in the figure, I am working with the nhanes database, I have grouped serum folate into quartiles to study the relationship between serum folate and diabetes, I used a univariate logistic analysis, but I found that in the final statistical results, the results of the global test are not statistically significant, but in the maximum likelihood ratio estimate, the fourth quartile is statistically significant compared to the first quartile. Why?

[PaigeMiller · Accepted solution]

Two different statistical tests, each testing different hypotheses, do not have to agree.

The overall test tests the hypothesis (in layman's words) that the model has found a statistically significant effect. In this case, the effect is not significant.

The individual tests looks to see if the coefficient on a single parameter (Serum group 4) is zero, or not. It is different from zero. But we already know the effect of serum group is not significant, in other words the different serum levels are not statistically different from each other. So despite serum group being statistically different than zero, the serums are not different from each other, which is a different hypothesis. No difference between the serums, and no significant intercept; thus model is not significant.