I think yes. This may be due to the weighted averaging process used in meta-analysis, which considers both the proportions and the sample sizes of each study.

Bishnu Deep Pathak, I meta-analyzed the 3 proportions you gave using this Stata code:

* Stata version 18 or later required.

clear

input study x n

1 10 10

2 24 24

3 54 54

end

meta esize x n, esize(logitprop)

meta summarize, transform(invlogit)

The pooled estimate is:

invlogit(theta) = 0.980, 95% CI: 0.906 to 0.996

PS- I forgot to say that these results are from a random effects model estimated via REML (the default settings in Stata).