gwern comments on LessWrong Survey Results: Do Ethical Theories Affect Behavior? - Less Wrong

18 Post author: peter_hurford 19 December 2012 08:40AM

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

Article Navigation

Comments (46)

Comments (46)

You are viewing a single comment's thread. Show more comments above.

Comment author: Unnamed 19 December 2012 11:44:47AM *  5 points [-]

Amounts given to charity have a highly skewed distribution, which makes it hard to find effects if you run analyses on the raw numbers. It can help to take logs, or to just look at categorical variables like giving anything vs. giving nothing.

I looked at the data using the second of those two approaches, seeing what percent of people donated any amount >0. I also combined moral views into two categories, consequentialists vs. everyone else (for simplicity & because of small sample sizes). The numbers:

"...have you donated to charity over the past year"
60.9% of consequentialists
60.7% of non-consequentialists
p = .95

"...have you donated to SIAI or CFAR in the past year"
15.1% of consequentialists
5.6% of non-consequentialists
p < .0001

"...have you donated to anti-aging related charities like SENS over the past year"
2.8% of consequentialists
0.6% of non-consequentialists
p = .02

So consequentialists are more likely than non-consequentialists to give to the weird ingroupy charities, but not more likely to give to charity overall. I played around with the data a bit in other ways (e.g., looking at log-giving, controlling for income) and this pattern seemed to hold up.

It's not clear if this relationship is causal; my guess is that it's (at least mostly) not. Consequentialism is associated with closer ties to the LW community on several measures, including sequence-reading, meetup attendance, and the other components of the composite LW exposure variable I've used elsewhere (karma, time in community, and LW use). Controlling statistically for LW exposure weakens the association between consequentialism & giving to SI/CFAR, and leaves it only marginally statistically significant (p = .053).

Consequentialism is also associated with various other elements of the local memeplex, including p(ManyWorlds), p(AntiAgathics), p(Simulation), and personal cryonics status/plans (but not statistically significantly with p(Cryonics)). Which suggests that it's not something about morality/donating in particular.

Comment author: gwern 19 December 2012 05:19:54PM 2 points [-]

Logging the total charity donations:

R> lw2012 <- read.csv("2012.csv")
R> consequentialism <- lw2012[as.integer(lw2012$MoralViews) == 2,]
R> consequentialism <- log1p(as.integer(as.character(consequentialism$Charity)))
R> consequentialism <- consequentialism[!is.na(consequentialism) & consequentialism != 0]
R> others <- lw2012[as.integer(lw2012$MoralViews) != 2,]
R> others <- log1p(as.integer(as.character(others$Charity)))
R> others <- others[!is.na(others) & others != 0]
R> t.test(consequentialism, others)
Welch Two Sample t-test
data: consequentialism and others
t = 1.687, df = 411.6, p-value = 0.09238
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-0.04613 0.60421
sample estimates:
mean of x mean of y
5.241 4.962
Comment author: Unnamed 20 December 2012 02:03:22AM 1 point [-]

It looks like you threw out the people who gave 0 to charity when you took the log. I typically use ln(x+1) for these types of variables, which maps zero to zero.

In this case, your approach leads to stronger effects (or at least lower p values). Repeating your analysis with the full data set (n=575), it's actually statistically significant on its own (Welch t=2.03, p=.043), means of 5.25 vs. 4.92. Excluding non-givers also gives lower p-values in the analyses I described here; when controlling just for age the effect of consequentialism is significant at p=.001 (geometric means $204 vs. $121). Even though the trend is for consequentialists to be more likely to give than non-consequentialists, it looks like the added noise of having all those points at zero has a bigger effect.

Comment author: gwern 20 December 2012 02:15:00AM 3 points [-]

I threw out zero because it seems like a non-response to me in a lot of questions. Someone giving a number shows they have at least responded with a non-default and did donate something, while someone leaving a zero may be simply equivalent to people who left empty responses.

(While I'm commenting on my analysis: from a utilitarian perspective, comparing logs doesn't even make sense; personal utility may follow some sort of logarithm in money, but charities don't - the problems are just too big for any one person.)

Powered by Reddit Powered by Reddit