1. Poll Thread: I've got to correct the misuse of the term "oversampled". Probably 99% of the time, pundits are using the wrong term.
E.g. "That poll is inaccurate BC it 'oversampled' this party, or this group of people."
"Look at how many more X there are."
Follow along...
E.g. "That poll is inaccurate BC it 'oversampled' this party, or this group of people."
"Look at how many more X there are."
Follow along...
2. That is an incorrect use of the term "oversampled", which means something altogether different.
What they really mean to say is that is certain group might very well be "overrepresented". That means the poll reflects a larger share of the population than is accurate.
What they really mean to say is that is certain group might very well be "overrepresented". That means the poll reflects a larger share of the population than is accurate.
3. What is "oversampling"? Best way to describe it is to use an example.
In Florida, we always "oversample" Hispanics. We ask a Latino detail question B/C of 1) there's a large number of Hispanic subgroups and 2) voting behavior for those groups differ, often significantly.
In Florida, we always "oversample" Hispanics. We ask a Latino detail question B/C of 1) there's a large number of Hispanic subgroups and 2) voting behavior for those groups differ, often significantly.
4. So, in order to accurately gauge the overall Hispanic population, we "oversample" to collect sufficient responses to reduce the sampling error for each subgroup.
A potentially unrealistic (non-representative) partisan breakdown is "overrepresentation".
Get the difference?
A potentially unrealistic (non-representative) partisan breakdown is "overrepresentation".
Get the difference?
5. Why is it important to know the difference?
One, the spreading of ignorance re: how polling is conducted spreads distrust.
Two, it could help to understand why a poll was or was not accurate. Overrepresentation will often tell us more about insufficient collection modes.
One, the spreading of ignorance re: how polling is conducted spreads distrust.
Two, it could help to understand why a poll was or was not accurate. Overrepresentation will often tell us more about insufficient collection modes.
6. Remember this the next time you hear cable pundits, see a tweet on Twitter, etc., saying a poll has "oversampled" Ds, Rs or whomever.
It's NOT accurate.
They may well have used a collection mode that led to "overrepresentation", but that's not the same as "oversampled".
It's NOT accurate.
They may well have used a collection mode that led to "overrepresentation", but that's not the same as "oversampled".
7. Need to add something else to answer a great Q.
Overrepresentation can be caused by a number of reasons, including aforementioned mode of collection, itself. But also very basic response bias, which is particularly a big problem for "gold standard" live-caller polls.
Overrepresentation can be caused by a number of reasons, including aforementioned mode of collection, itself. But also very basic response bias, which is particularly a big problem for "gold standard" live-caller polls.
8. Perhaps something went wrong with weighting. When weighting variables are applied, it will change the breakdown of other groups, to include party or other demographics.
In 2016, weighting properly for region and education—or not at all—were a major issue for big media polls.
In 2016, weighting properly for region and education—or not at all—were a major issue for big media polls.
