Phil Swatton
@philswatton.bsky.social
Work as a data scientist at the Alan Turing Institute, background in political science. Views my own and not necessarily shared by my employer.
https://philswatton.github.io/
https://philswatton.github.io/
I tried to make a case along these lines not too long ago: renewal.org.uk/blog/public-...
Public Opinion and the Survey Response
Political scientists know that survey responses cannot be taken at face value. For most respondents, most of the time, responses are not the product of deeply held, pre-existing, coherent views on par...
renewal.org.uk
November 14, 2025 at 12:22 PM
I tried to make a case along these lines not too long ago: renewal.org.uk/blog/public-...
That makes a lot of sense, thank you!
November 12, 2025 at 2:11 PM
That makes a lot of sense, thank you!
Wilcoxon is an approach to it, but I guess my q is:
- If tests on individual datasets fail to reject the null (can't tell which is better on any given dataset)
- & a single test comparing accuracies across datasets rejects the null (A is better than B result across datasets)
What should I infer?
- If tests on individual datasets fail to reject the null (can't tell which is better on any given dataset)
- & a single test comparing accuracies across datasets rejects the null (A is better than B result across datasets)
What should I infer?
November 12, 2025 at 2:03 PM
Wilcoxon is an approach to it, but I guess my q is:
- If tests on individual datasets fail to reject the null (can't tell which is better on any given dataset)
- & a single test comparing accuracies across datasets rejects the null (A is better than B result across datasets)
What should I infer?
- If tests on individual datasets fail to reject the null (can't tell which is better on any given dataset)
- & a single test comparing accuracies across datasets rejects the null (A is better than B result across datasets)
What should I infer?
Paper in the brackets I forgot to link is www.jmlr.org/papers/volum...
www.jmlr.org
November 12, 2025 at 1:42 PM
Paper in the brackets I forgot to link is www.jmlr.org/papers/volum...
Do you:
1) interpret the results as being inconclusive on which classifier is better
2) interpret classifier A as being better than B
Obviously there is
3) Find some extra datasets w/ larger test sets, but I'm curious how people would approach the initial problem
(3/3)
1) interpret the results as being inconclusive on which classifier is better
2) interpret classifier A as being better than B
Obviously there is
3) Find some extra datasets w/ larger test sets, but I'm curious how people would approach the initial problem
(3/3)
November 12, 2025 at 1:40 PM
Do you:
1) interpret the results as being inconclusive on which classifier is better
2) interpret classifier A as being better than B
Obviously there is
3) Find some extra datasets w/ larger test sets, but I'm curious how people would approach the initial problem
(3/3)
1) interpret the results as being inconclusive on which classifier is better
2) interpret classifier A as being better than B
Obviously there is
3) Find some extra datasets w/ larger test sets, but I'm curious how people would approach the initial problem
(3/3)
Classifier A has consistently better accuracy than classifier B on most test sets (say, 13/15). This is significant in a Wilcoxon signed rank test (approach advocated by ).
However, on most _individual_ points (say, 14/15), the 95% CIs on the accuracy on each dataset overlap. (2/3)
However, on most _individual_ points (say, 14/15), the 95% CIs on the accuracy on each dataset overlap. (2/3)
November 12, 2025 at 1:40 PM
Classifier A has consistently better accuracy than classifier B on most test sets (say, 13/15). This is significant in a Wilcoxon signed rank test (approach advocated by ).
However, on most _individual_ points (say, 14/15), the 95% CIs on the accuracy on each dataset overlap. (2/3)
However, on most _individual_ points (say, 14/15), the 95% CIs on the accuracy on each dataset overlap. (2/3)
Hi Ben, I sent an email about the role but fear it may have ended up in your spam
November 7, 2025 at 6:18 PM
Hi Ben, I sent an email about the role but fear it may have ended up in your spam
And good shout on DKs/would not votes - will try that out later today if I remember!
November 7, 2025 at 5:32 PM
And good shout on DKs/would not votes - will try that out later today if I remember!
If I squint hard enough, I might interpret it as established party vs not, in that it explains both the gap between Con & Ref on the one hand, and why Green voters are on the same part as Ref on the other, but I'm not fully convinced by it - e.g. why Lab & LD are more middling on the dimension
November 7, 2025 at 5:32 PM
If I squint hard enough, I might interpret it as established party vs not, in that it explains both the gap between Con & Ref on the one hand, and why Green voters are on the same part as Ref on the other, but I'm not fully convinced by it - e.g. why Lab & LD are more middling on the dimension
Thank you - the paper looks fascinating, will add to my reading pile, thanks for sharing!
November 7, 2025 at 12:08 PM
Thank you - the paper looks fascinating, will add to my reading pile, thanks for sharing!
Based on our recent discussion, maybe or maybe not of interest to
@mariosrichards.bsky.social
@ralphscott.bsky.social
@jack-bailey.co.uk
@heinzbrandenburg.bsky.social
@mariosrichards.bsky.social
@ralphscott.bsky.social
@jack-bailey.co.uk
@heinzbrandenburg.bsky.social
November 7, 2025 at 11:57 AM
Based on our recent discussion, maybe or maybe not of interest to
@mariosrichards.bsky.social
@ralphscott.bsky.social
@jack-bailey.co.uk
@heinzbrandenburg.bsky.social
@mariosrichards.bsky.social
@ralphscott.bsky.social
@jack-bailey.co.uk
@heinzbrandenburg.bsky.social
I've made the code for the ANES output available at: gist.github.com/philswatton/...
gist.github.com
November 7, 2025 at 11:55 AM
I've made the code for the ANES output available at: gist.github.com/philswatton/...
(this is for most but not all post-election feeling thermometers)
November 4, 2025 at 9:49 PM
(this is for most but not all post-election feeling thermometers)
Here's the equivalent distributions for ANES 2024, they look much spikier (but possibly you still get more out of it, if e.g. people use values ending 0 or 5, that's still 21 meaningful values vs 11 in 0-10 or 7 in 1-7)
November 4, 2025 at 9:49 PM
Here's the equivalent distributions for ANES 2024, they look much spikier (but possibly you still get more out of it, if e.g. people use values ending 0 or 5, that's still 21 meaningful values vs 11 in 0-10 or 7 in 1-7)
Made this example w/ the 2024 data, will upload a blog post or gist over the next couple of days:
November 4, 2025 at 9:42 PM
Made this example w/ the 2024 data, will upload a blog post or gist over the next couple of days:
That's really interesting - what dataset is this from?
November 4, 2025 at 6:11 PM
That's really interesting - what dataset is this from?
Agreed - I'd be interested in seeing comparisons of 1-7, 0-10, and 0-100 disaggregated across ideological self placements vs warmth ratings too
November 4, 2025 at 5:43 PM
Agreed - I'd be interested in seeing comparisons of 1-7, 0-10, and 0-100 disaggregated across ideological self placements vs warmth ratings too
1-7 scales are on issues
0-100 are warmth, but you can recover a liberal-conservative dimension from them (I think I have a better plot somewhere):
0-100 are warmth, but you can recover a liberal-conservative dimension from them (I think I have a better plot somewhere):
November 4, 2025 at 5:33 PM
1-7 scales are on issues
0-100 are warmth, but you can recover a liberal-conservative dimension from them (I think I have a better plot somewhere):
0-100 are warmth, but you can recover a liberal-conservative dimension from them (I think I have a better plot somewhere):
I don't know of any dataset that would enable comparison of different scales for warmth ratings though
November 4, 2025 at 5:31 PM
I don't know of any dataset that would enable comparison of different scales for warmth ratings though
It's been a while and I don't have my laptop with me to check atm, but while I imagine a lot of respondents do use '50', '25', etc I think I recall correctly that there was enough variation between different stimuli to rate
November 4, 2025 at 5:31 PM
It's been a while and I don't have my laptop with me to check atm, but while I imagine a lot of respondents do use '50', '25', etc I think I recall correctly that there was enough variation between different stimuli to rate
I'm not sure actually - I wouldn't use 0-100 for issue scales - when I wish for them it's more about having warmth ratings towards lots of different stimuli. I helped make a presentation to the RSS on the 2020 presidential w/ them, they were really interesting for nonmetric unfolding:
November 4, 2025 at 5:31 PM
I'm not sure actually - I wouldn't use 0-100 for issue scales - when I wish for them it's more about having warmth ratings towards lots of different stimuli. I helped make a presentation to the RSS on the 2020 presidential w/ them, they were really interesting for nonmetric unfolding: