« first day (2188 days earlier)   

2:04 PM
Can someone please have a look at this answer? skeptics.stackexchange.com/a/37340/1044 ? I'm confused as to whether it's original research or not; seems good enough as to not warrant a flag.
 
 
7 hours later…
Joe
9:10 PM
@Sklivvz Hi - can I ask what specifically you consider theoretical about my answer?
I'm not intending it to be a research answer or a math answer, but pointing out that the article misused the poll's data
Is there a way to do that while still staying on-topic for Skeptics?
I did cite one source (the census numbers) showing the numbers in the poll clearly do not represent a valid sample in the way the Times was using them; the rest of the answer is explaining why/how/etc., but if that part is inappropriate but the first part is okay, I can remove it
 
9:23 PM
@Joe well the answer is basically a lot of guesswork, but most of it is just something you are asserting without actual evidence. It's OK to point out evident flaws in a paper, but I think a lot of what your answer states is a matter of interpretation, and as such it needs at the very least some specialist validation.
 
Joe
@Sklivvz The problem is that all you can do is guess at the methodology
(and I am a specialist in this area, fwiw, though I know that's not particularly relevant)
you can show (through the census data I cite) that their proportions do not match the population
 
I understand, and I guessed so, that's evident from what your write
 
Joe
Beyond that, it's guesswork as to why they don't match
so if that's not appropriate, I can remove it, though I feel like for someone who doesn't know the field, it's helpful to understand the different possible ways it could be wrong
 
One way we've used in the past to validate such guesswork is to ask a parallel question on a specialist stackexchange and use their experts
 
Joe
The main answer on that question though I feel is not very good, because it takes as given the validity of the poll's data and the Times' interpretation of such
which is simply wrong - the Times does something that is simply wrong to improperly use the demographic data the way they do
 
9:27 PM
Right, right, I understand. Do you think that we could ask crossvalidated.se to take a look at the paper and use that to support your answer?
 
Joe
I certainly wouldn't object to that, though I don't know any survey researchers over there
I assume there must be some
(this isn't actually a statistics questions per se, it's a survey research question, but you're not going to find an SE on that specifically I don't think...)
 
mmm, let me check
 
Joe
I might also be able to find a source or two that discuss this kind of error
(it's a common one in survey research)
 
That'd be good, we could restrict the answer to the specific error and that would fix it. However, I'd rather keep the other part as well if possible
 
Joe
Well, the idea would be to cover the other part as part of that I think
We can trivially show the numbers are wrong (the census supports that directly, imo, plus the issue JimmyJames noted)
and then explain why through the other source
would that make sense?
 
9:33 PM
If you can clearly distinguish whatever we have evidence for (e.g. a poll with a recognised similar error would be good evidence that this poll has the same issue) and what is only a possible explanation, then it would be much better
But, I'm convinced that a lot of your guesswork is actually a correct explanation so I'm trying to think how to strengthen that part...
crossvalidated would probably be the best shot for asking for more eyes. the other possibility is politics but... it's not scientifically strong.
 
Joe
Okay, if you manage to find someone over there that would be great
I will look around for articles on the kind of error... it's hard to search for sadly
"error" means something very different in survey research
 
yep, i understand
 

« first day (2188 days earlier)