The overall average hides 10 different numbers

city splits data-job-postings into 10 groups. Here is what salary_min_usd looks like inside each.

• San Francisco — mean 132,017, median 130,500 (532 rows)
• New York — mean 126,626, median 129,500 (495 rows)
• Austin — mean 104,188, median 105,500 (253 rows)
• Chicago — mean 101,232, median 102,500 (299 rows)
• London — mean 64,365, median 69,000 (436 rows)
• Amsterdam — mean 62,590, median 69,500 (245 rows)

Top to bottom that is 132,017 against 25,038, a spread of 427.3%. The pooled average is 104,106.

select city,
  count(*)                                                    as rows,
  round(avg(salary_min_usd)::numeric, 2)                      as mean,
  percentile_cont(0.5) within group (order by salary_min_usd) as median
from data_job_postings
group by 1
order by mean desc;

A spread that wide means the pooled number is not a summary, it is an artefact of the mix. Change the proportion of San Francisco rows and the overall average moves without any individual group changing at all — which is how a metric goes up while every segment goes down.

Notice the mean and median columns disagree only slightly here. Always compute both in the group-by. The comparison between them per segment is free and tells you whether you are looking at a level difference or a tail difference.

This is the setup for Simpson's paradox — the case where every segment moves one way and the total moves the other.