Preregistration Of Hypotheses For The SSC Survey

[This post is about the 2018 SSC Survey. If you’ve read at least one blog post here before, please take the survey if you haven’t already. Please don’t read on until you’ve taken it, since this could bias your results.]

I’m preregistering my hypotheses for the survey this year. So far I’ve glanced at Google’s bar graphs for each individual question but haven’t started exploring relationships yet, so I’m not cheating too badly. I’ll still look for things I haven’t preregistered, but I’ll admit they’re preliminary results only. This is the stuff I’ve been thinking about beforehand and will be taking more seriously:

1. I plan to replicate the general thrust of last year’s results reported in Can We Link Perception And Cognition on the sample of new people who didn’t take the survey last year. In particular, I’m expecting that weirder, more autistic, more liberal, more schizophrenic, and more transgender people will be more likely to display unusual patterns of perception (hollowness or ambiguity) in the Hollow Mask illusion. I expect this to become much more obvious since I’ve included three examples of the illusion this year including one that seems to give a wider diversity of results.

1a. I plan to replicate the results from last year that people who were better at noticing duplicate “the’s” are more likely to display unusual patterns of perception on the Hollow Mask illusion.

2. I plan to conceptually replicate Mitchell et al’s study showing that autistic people are less susceptible to the Shepherd Table Illusion.

3. I plan to conceptually replicate Caparos et al’s study showing that politically further-right people are more likely to use global processing on a Navon task (eg when there’s an H made of tiny Es, they see the H more than the Es).

4. I plan to investigate a general construct of “first sight and second thoughts” that involves people being better able to see what’s actually there, and less susceptible to illusions, priors, stereotypes, and assumptions. This will involve correlations between the two Duplicate Thes illusions, the Hollow Mask illusion, the Shepherd Table illusion, the Cookies illusion, the Parentheses palindrome, the Map riddle, the Surgeon riddle, the Switched Answers task, the Cognitive Reflection test, and the Wason task.

4a. If I can figure out how to get a common factor out of all of these, I plan to see if it’s the same thing I’m looking at in 1, and how it relates to the same groups.

4b. Whether this relates to a general willingness to believe strange or unpopular things. Check vs. AI risk concern and HBD support.

5. I plan to investigate a general construct of “ambiguity tolerance” that involves people being okay with a superposition of different conflicting ideas. This will involve correlations between ambiguous results on the Hollow Mask illusion, the Spinning Dancer illusion and the Squares-Circles illusion, and with answers to the questions from the Tolerance Of Ambiguity and Tolerance of Uncertainty scales.

5a. Whether perceptual ambiguity relates to cognitive ambiguity. I want to check whether people with high ambiguity tolerance on the optical illusions are more likely to say their political opponents have some good points, are less likely to say their political opponents are evil, and are less likely to say the existing political system is justifiable. Also if they’re more likely to enjoy puns.

5b. To what degree this is the same construct as (1), and is stronger among the same demographic groups.

5c. I also want to see if people with high ambiguity tolerance give less extreme answers on questions in general. I’ll probably use Ambition, Social Status, Romantic Life, and Morality for this, just because these seem like complicated questions there’s no obvious right answer to.

5d. I plan to confirm previous studies showing low ambiguity tolerance correlates with conservative philosophy; check vs. Political Spectrum 1-10. I predict that this will be stronger for populists than for “business conservatives”, so I expect the low ambiguity correlation will be weak for generic conservatives, stronger for Trump supporters, strongest for people who identify as alt-right.

6. I plan to investigate whether autistic people are more likely to give process-centered rather than person-centered answers to the two political categorization questions (categorizing Nazis, categorizing civil disobedience on gay marriage). That is, neurotypical people will be more likely to categorize based on which side wins, and autistic people will be more likely to categorize based on what procedures were followed (eg violence, civil disobedience).

6a. I also want to investigate how these correlate with political views. I may end up controlling for this as a confounder in (6) above.

6b. This is a totally wild out-of-left field idea, but I suppose I should check how these relate to the Navon figures since they’re both about categorization.

7. I plan to confirm or disprove, once and for all, whether our community has more older siblings. For lack of a fancier way to do this, I’ll take the set of all people who have exactly one sibling, and see what percent of them are older vs. younger. If it’s significantly above 50% older, I’m going to interpret this as a birth order effect. I’ll do the same with the set of people who have two siblings, three siblings, etc, and combine them all for a final determination. Half-siblings will be ignored. If you have any problems with this methodology, tell me now.

7a. If I find we’re disproportionately older, try to use subgroups to figure out where the effect is stronger or weaker, to try to find exactly what’s going on. For example, are Less Wrongers more older-skewed than SSC readers in general?

7b. Birth order by autism, Openness, and IQ/SAT.

7c. One traditional birth-order claim is that younger children are more rebellious, so check birth order vs. people who think system needs to be fine-tuned or destroyed.

8. I plan to conceptually replicate studies showing that the more older brothers (but not younger brothers, or older or younger sisters) you have, the more likely you are to be gay.

8b. See if this predicts anything else: bisexuality, transgender, gender non-conformity, political leftism, autism, possibly ‘first sight and second thoughts’, possibly ‘ambiguity tolerance’.

9. I plan to see whether people with ADHD are more likely to prefer the buzzing city aesthetic to the quiet village aesthetic, more likely to rate themselves as more risk-taking, and more likely to describe themselves as ambitious.

10. I plan to investigate the hypothesis about sexual harassment mentioned here: that it’s higher in gender imbalanced industries only due to potential-perpetrator-to-victim ratio. I predict that in relatively gender imbalanced industries (in terms of survey categories, all three Computers fields, Finance, Physics, and Mathematics) compared to relatively gender-balanced industries (Health Care, Psychology, Art, Law, Biology), a higher percent of women will report being harassed at work, but the percent of men reporting harassing at work will remain the same.

10b. I predict that the more people identify with social justice, and the more positively they feel about feminism, the more likely they are to report both being harassed and harassing others, due to more awareness and lower threshold to report. I predict poor social skills and autism spectrum will predict more likely to say one is a harasser, due to causing unintentional offense. I predict people who are harassed more at work will also be harassed more outside of work.

11. A long time ago, I randomized people into groups and made them read articles on AI risk to see how it changed their minds. The effect mostly persisted after one month. Since those groups were randomized by birth date, and I asked respondents their birthdates, I plan to see if those effects continue to persist after a year.

These are mostly conceptual descriptions of what I’m going to do rather than algorithmic descriptions of exactly how I’m going to process the data. Part of that is that a lot of this involves statistical techniques at the limits of my abilities and I’m going to have to see if I can actually do them. Most important, I would like to learn enough about factor analysis to actually check for a General Factor Of First Sight/Second Thoughts, and a General Factor of Ambiguity Tolerance. If I have them, I’d like to use them to see if they correlate with the other things I’m wondering if they correlate with. If I can’t make this work or beg someone else to do it for me, I’ll just eyeball the correlations between individual questions, see which ones are highest, and maybe take an average of those questions or something.

Mostly I won’t be doing anything fancy or with too many branching paths to the data, but I plan to operationalize autism in two ways. First, a scale where professional diagnosis equals 3, self-diagnosis equals 2, family member equals 1, and no personal/family history equals 0. Second, the Autism Spectrum Quotient test I made people take at the bottom of the survey. I’m not at all confident these will correlate more than a weak amount, but I’ll try it and see. I might also try some kind of average of the two measures. Since there are a few things I expect to be correlated with autism – mathematical careers, bad response to clothing tags, poor social skills – I might check to see whether the first measure, the second measure, or the combination does a better job of predicting these, and stick with whichever one does. I’ll try not to base which measure I use on any of the variables I’m actually testing.