One way is to let stats count by type build the potentially incomplete set, then append a set of "dummy" rows on the end where each row has a count of "0", then do a stats sum(count) as count at the end.
<<your search terms here>>
| stats count by type
| append [| stats count | fields - count | eval type=split("MissingA,MissingB,MissingC,MissingD",",") | mvexpand type | eval count=0]
| stats sum(count) as count by type
This will give a table that looks like this, and that includes "0" rows even when the given type is missing entirely from the raw data.
------------------------------------
| type | count |
| MissingA | 12 |
| MissingB | 0 |
| MissingC | 3 |
| MissingD | 0 |
There's another way that you might find easier, or that you might find interesting. That's to create a "chart style" set of output rows first, use a simple fillnull command to add your "0" entries if they're missing (instead of that append command), then use untable to turn the "chart style" output into "stats style" rows like the above.
<<your search terms here>>
| eval foo="1"
| chart count over foo by type
| fillnull MissingA MissingB MissingC MissingD value="0"
| fields - NULL
| untable foo type count
| fields - foo
The benefit of this approach is that there's no append. If you have moral/ethical/religious objections to append (not unreasonable), then this approach may be preferred!
Admittedly the little "foo" trick is clunky and funny looking. untable and xyseries don't have a way of working with only 2 fields so as a workaround you have to give it a dummy third field and then take it away at the end.
