我有一个表social_accounts，facebook_id其中列上有部分索引user_id IS NULL。

如果我执行简单查询WHERE facebook_id = '123'，则使用索引：

 => EXPLAIN for: SELECT "social_accounts".* FROM "social_accounts"  WHERE (user_id IS NOT NULL) AND "social_accounts"."facebook_id" = '123'
                                                  QUERY PLAN
--------------------------------------------------------------------------------------------------------------
 Index Scan using index_social_accounts_on_facebook_id on social_accounts  (cost=0.00..8.28 rows=1 width=345)
   Index Cond: ((facebook_id)::text = '123'::text)
   Filter: (user_id IS NOT NULL)

但是如果我使用IN它进行查询，则不会使用索引：

 => EXPLAIN for: SELECT "social_accounts".* FROM "social_accounts"  WHERE (user_id IS NOT NULL) AND "social_accounts"."facebook_id" IN ('123', '456')
                                            QUERY PLAN
---------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on social_accounts  (cost=8.53..16.36 rows=2 width=345)
   Recheck Cond: ((facebook_id)::text = ANY ('{123,456}'::text[]))
   Filter: (user_id IS NOT NULL)
   ->  Bitmap Index Scan on index_social_accounts_on_facebook_id  (cost=0.00..8.52 rows=2 width=0)
         Index Cond: ((facebook_id)::text = ANY ('{123,456}'::text[]))
(5 rows)

为什么在第二种情况下不使用索引？有什么办法可以加快这个查询？

（请注意，在此示例中，我已截断了数组，并测试了更多元素，但结果却相同，缓慢）