Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

gwern comments on Open thread, Sep. 26 - Oct. 02, 2016 - Less Wrong Discussion

2 Post author: MrMind 26 September 2016 07:41AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (90)

You are viewing a single comment's thread.

Comment author: gwern 28 September 2016 11:09:54PM 2 points [-]

Continuing my catnip research, I'm preparing to run a survey on gwern.net & Mechanical Turk about catnip responses. I have a draft survey done and would appreciate any feedback about brokenness or confusing questions: https://docs.google.com/forms/d/e/1FAIpQLSeT3GIg-pSwzDFAfNaqE-MzfJEtD0HghN_Vma68OZJtz1Pztg/viewform

Comment author: gwern 29 September 2016 11:34:19PM *  0 points [-]

OK, no complaints so far, so I'm just going to launch it. Consider the survey now live. Did I mention that there will be cake?

Comment author: Elo 30 September 2016 12:50:31AM -2 points [-]

cat weight might be relevant, cat current age, cat body shape (fat/skinny), description of cat's response to catnip,

Comment author: Elo 30 September 2016 12:48:06AM -2 points [-]

I am no expert, but I wonder if you could run a monte-carlo on your expected responses. Do the questions you ask give you enough information to yield results?

Just not sure if your questions are honing correctly. Chances are there are people that know better than me.

Comment author: gwern 30 September 2016 01:06:17AM *  0 points [-]

If I get at least 100 responses, then that will help narrow down the primary question of overall catnip response rate adequately in combination with the existing meta-analysis. I expect to get at least that many, and in the worst case I do not, I will simply buy the survey responses on Mechanical Turk.

The secondary question, Japanese/Australian catnip rates vs the rest of the world, I do not expect to get enough responses since the power analysis of the 60% vs 90% (the current average vs Japanese estimates) says I need at least 33 Japanese respondents for the basic comparison; however, Mechanical Turk allows you to limit workers by country, so my plan is to, once I see how many responses I get to the regular survey, launch country-limited surveys to get the necessary sample size. I can get ~165 survey responses with a decent per-worker reward for ~\$108, so split over Japan/Korea/Australia, that ought to be adequate for the cross-country comparisons. (Japan, because that's where the anomaly is; Korea, to see if the anomaly might be due to a bottleneck in the transmission of cats from Korea to Japan back in 600-1000 CE; Australia, because a guy on Twitter told me Australian cats have very high catnip response rates; and I hopefully will get enough American/etc country responses to not need to pay for more Turk samples from other countries.) Of course, if the results are ambiguous, I will simply collect more data, as I'm under no time limits or anything.

For the tertiary question, response rates to silvervine/etc, I am not sure that it is feasible to do surveys on them. There is not much mention of them online compared to catnip, and they can be hard to get. My best guess is that of the cat owners who have used catnip, <5% of them have ever tried anything else, in which case even if I get 200 responses, I'll only have 25 responses covering the others, which will give very imprecise estimates and not allow for any sort of modeling of response rates conditional on being catnip immune or factor analysis. If I'm right and the survey is unable to answer the question without recruiting thousands of cat owners, then that tells me I will have no choice but to continue experimenting myself and contact the local pound & animal rescue organizations asking if I can use my battery of substances on their sets of cats.

As for your question suggestions: weight/current-age/body-shape-fatness haven't been suggested in the catnip literature as moderators, current age seems like it should be irrelevant, and asking for a free response description of the catnip response is really burdensome on the user compared to multiple-choice or checkboxes (survey guidelines emphasize as few free-responses as possible, no more than 1 or 2) and the catnip response is pretty stereotypical even across species so there wouldn't be much point.

Comment author: Lumifer 30 September 2016 02:38:27PM 0 points [-]

Did I mention that there will be cake?