The row size is much bigger for your second try: `width=157` vs. `width=46`. Postgres will even *more readily* use an index for wider rows. Possible reasons for the unexpected sequential scan include:

- You have substantially fewer rows in your tables for the second test as indicated by planner estimates: `rows=143` vs. `rows=357`. It does not pay to look up an index for only few rows to sort.

- Or statistics are outdated leading to misguided planner estimates (Postgres only *thinks* there would be fewer rows).

- The index size may have been bloated as a side-effect of rewriting the tables. `REINDEX` or `VACUUM FULL` would repair that.

Run `VACUUM FULL ANALYZE` on all involved tables and try again - with the same number of rows in all tables. You should see bitmap index scans again. If the phenomenon persists, provide the output of `EXPLAIN (ANALYZE, BUFFERS)`, not just `EXPLAIN`.

This [**SQL fiddle**][1] with 10k rows per child and valid statistics shows bitmap index scans as expected.

About testing indexes SQL Fiddle:

- http://dba.stackexchange.com/questions/99323/postgresql-partial-index-unused-when-created-on-a-table-with-existing-data/99324#99324


  [1]: http://sqlfiddle.com/#!15/3aa47/1