MNF-I has provided their methods for measuring ethno-sectarian violence in Iraq to Talking Points Memo. The document makes it clear that deaths from bombings are supposed to be counted, so why didn't they count the over five hundred Yazidi dead in Qahtaniya. These deaths should have been counted, since the definition is
Ethno-sectarian violence is defined as an event and any associated civilian deaths caused by ... and all types of explosive devices identified as being conducted by one ethnic/religious person/group directed at a different ethnic/religious person/group, where the primary motivation for the event is identified as ethnic or religious
That is clear enough to show that deaths from the Qahtaniya bombings should have been counted. However all of the examples refer to Shi'ites killing Sunnis or vice versa. There is no mention of possible ethnic violence and no mention of religions other than Islam.
I assume that over at MNF_I they decided that Yazidi don't count, because counting them would make the August numbers look as awful as the reality. The methodology as described was clearly not applied.
Spencer Ackerman explains how the trends are so different from civilian deaths rates reported by independent sources
For executions, murders and kidnappings -- situations in which sectarianism may be difficult to determine -- MNF-I says it uses "host nation" reporting in addition to its own. Many media and non-governmental organizations consider information on casualties released by the Iraqi ministries to be self-serving, misleading or contradictory.
Basically the US military decided that the ethnic cleansers were a reliable source on the level of ethnic cleansing. This is nonsense as was shown by you guessed it Talking Points Memo. [TPM
I linked to you when you didn't link to yourselves. not that that's a *hint* or anything]
Notice the alleged improvement occurred just when Iraqi authorities stopped allowing the UN to see the raw data.
update: typo corrected.