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.
I realize that I didn’t do a very thorough job of looking at total charitable giving, so I did some more analyses.
Summary: People tend to give more if they are older, richer, or more religious. Consequentialists tend to be younger, but after controlling for age they do tend to give more than non-consequentialists (p=.02). Consequentialists also tend to be less religious, and if you control statistically for that as well then the relationship between consequentialism and giving is even stronger (p=.006).
log(charity+1) seems like the best variable to look at for total charitable giving—it has roughly a normal distribution plus a big point mass off in the left tail at zero giving (which stays at zero after adding 1 & taking the log), n=952. There is a weak & nonsignificant trend for consequentialists to have higher log(charity+1), p=.30. Average log(charity+1) is 3.21 vs. 3.01 for the 2 groups, which means that the geometric means are e^3.21 = $25 vs. e^3.01 = $20, a 1.2x ratio.
There are other variables which have stronger, clearer relationships with total charitable giving: income, age, and religion. People with more money give more, older people give more, and more religious people give more. All three variables have independent effects—in a multiple linear regression, age, income, and religiosity are all significant predictors of charitable giving.
(Details on those effects: for income, I used log(income+1000). About a third of the data points are lost with analyses that include income because a lot of people left it blank, and that’s a non-random subset as they have lower giving than those who reported their income. The religion effect shows up on p(God) and p(Religion), and on the religious views question treated as categories, and on the religious views question treated as a quantitative scale from 1 (atheist & not spiritual) to 6 (committed theist). I combined the 3 quantitative questions into a single religiosity scale by standardizing each & averaging them.)
What does this tell us about the relationship between consequentialism and giving? Well, consequentialism is negative correlated with religiosity (consequentialists are less religious) and age (consequentialists are younger), which could hide a positive effect of consequentialism on giving. (It is uncorrelated with income).
And in fact, in a multiple regression using all four variables as predictors (age, income, religiosity, and consequentialism), all four are statistically significant predictors of charitable giving. The effect of consequentialism is significant at p=.006, and the effect size suggests that consequentialists give 1.9x as much as nonconsequentialists (of the same age, income, and religiosity); $42 vs. $22 geometric means (these are higher than the means before, since it excludes the people who left income blank, who tended to give less). So consequentialism does predict higher giving, controlling for age, income, and religiosity.
It’s not clear if we should be controlling for religiosity in this way, though—if someone buys into the idea package which includes atheism & consequentialism then maybe they shouldn’t get credit for being more generous than those who accept atheism but not consequentialism (especially when other idea packages, like theistic ones, are associated with higher generosity). But controlling only for age & income still leaves the effect of consequentialism statistically significant, p=.02, with consequentialists giving 1.7x as much as non-consequentialists of the same age & income ($40 vs. $23).
There is also some concern with controlling for income, especially because of the missing data (from a non-random third of the sample). There are also possible issues with endogeneity (e.g., people who choose a high income job so they can give more money to charity), although that suggests that controlling for income could lead to understating the effect of consequentialism rather than to overstating it. But it turns out to not be worth worrying about; running an analysis which controls only for age, consequentialism is still a significant predictor of giving, p=.02, 1.6x the giving ($27 vs. $17).
And there aren’t similar concerns about statistically controlling for age, since age is an exogenous variable—the causal arrows can only go in one direction there. Another way to account for age is to just exclude everyone under the age of 25 (since many people in that age range aren’t financially independent, so their giving rates aren’t that informative). Of those aged 25+, consequentialists give more than non-consequentialists, p = .04, 1.7x the giving ($56 vs. $32), n = 526.
One more analysis, inspired in part by Gwern’s comment here.
In my first analysis I broke people into two groups using a cutoff at the bottom of the distribution, giving anything vs. giving nothing. But if we care about how much money charities receive, then we should care more about the top of the distribution, because that’s where most of the money is coming from (80-20 rule and all).
So, let’s focus on the top of the distribution by putting a cutoff up there. $1000 is an especially convenient Schelling point, since it is the 80-20 point among donors: 21.8% of those who gave to charity gave $1000+, and they accounted for 81.4% of the donations.
1164 people answered the question about moral views, and 10.8% of those reported giving $1000 or more to charity (this counts those who left the charity question blank in the under-$1000 group). Breaking that down by consequentialism:
12.5% of consequentialists gave $1000+ 7.7% of non-consequentialists gave $1000+ p = .01.
Earlier, I found that consequentialists and non-consequentialists are equally likely to give to charity (vs. giving nothing). But here, we see that consequentialists are more likely to be big-money donors ($1000+).
Age, income, and religiosity are also significantly predictive of giving $1000+, and consequentialism remains a significant predictor (p=.002) after controlling for them.
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.
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.)
Interestingly for the memeplex theory (which seems very plausible), Consequentialists are also more likely to indicate understanding of the Newcombs Problem (by choosing one box or two box over not sure or no answer) [chi2 = 16.1393, p = 0.001] but not more likely to actually one box over two box (chi2 = 4.1755, p = 0.243).
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.
I realize that I didn’t do a very thorough job of looking at total charitable giving, so I did some more analyses.
Summary: People tend to give more if they are older, richer, or more religious. Consequentialists tend to be younger, but after controlling for age they do tend to give more than non-consequentialists (p=.02). Consequentialists also tend to be less religious, and if you control statistically for that as well then the relationship between consequentialism and giving is even stronger (p=.006).
log(charity+1) seems like the best variable to look at for total charitable giving—it has roughly a normal distribution plus a big point mass off in the left tail at zero giving (which stays at zero after adding 1 & taking the log), n=952. There is a weak & nonsignificant trend for consequentialists to have higher log(charity+1), p=.30. Average log(charity+1) is 3.21 vs. 3.01 for the 2 groups, which means that the geometric means are e^3.21 = $25 vs. e^3.01 = $20, a 1.2x ratio.
There are other variables which have stronger, clearer relationships with total charitable giving: income, age, and religion. People with more money give more, older people give more, and more religious people give more. All three variables have independent effects—in a multiple linear regression, age, income, and religiosity are all significant predictors of charitable giving.
(Details on those effects: for income, I used log(income+1000). About a third of the data points are lost with analyses that include income because a lot of people left it blank, and that’s a non-random subset as they have lower giving than those who reported their income. The religion effect shows up on p(God) and p(Religion), and on the religious views question treated as categories, and on the religious views question treated as a quantitative scale from 1 (atheist & not spiritual) to 6 (committed theist). I combined the 3 quantitative questions into a single religiosity scale by standardizing each & averaging them.)
What does this tell us about the relationship between consequentialism and giving? Well, consequentialism is negative correlated with religiosity (consequentialists are less religious) and age (consequentialists are younger), which could hide a positive effect of consequentialism on giving. (It is uncorrelated with income).
And in fact, in a multiple regression using all four variables as predictors (age, income, religiosity, and consequentialism), all four are statistically significant predictors of charitable giving. The effect of consequentialism is significant at p=.006, and the effect size suggests that consequentialists give 1.9x as much as nonconsequentialists (of the same age, income, and religiosity); $42 vs. $22 geometric means (these are higher than the means before, since it excludes the people who left income blank, who tended to give less). So consequentialism does predict higher giving, controlling for age, income, and religiosity.
It’s not clear if we should be controlling for religiosity in this way, though—if someone buys into the idea package which includes atheism & consequentialism then maybe they shouldn’t get credit for being more generous than those who accept atheism but not consequentialism (especially when other idea packages, like theistic ones, are associated with higher generosity). But controlling only for age & income still leaves the effect of consequentialism statistically significant, p=.02, with consequentialists giving 1.7x as much as non-consequentialists of the same age & income ($40 vs. $23).
There is also some concern with controlling for income, especially because of the missing data (from a non-random third of the sample). There are also possible issues with endogeneity (e.g., people who choose a high income job so they can give more money to charity), although that suggests that controlling for income could lead to understating the effect of consequentialism rather than to overstating it. But it turns out to not be worth worrying about; running an analysis which controls only for age, consequentialism is still a significant predictor of giving, p=.02, 1.6x the giving ($27 vs. $17).
And there aren’t similar concerns about statistically controlling for age, since age is an exogenous variable—the causal arrows can only go in one direction there. Another way to account for age is to just exclude everyone under the age of 25 (since many people in that age range aren’t financially independent, so their giving rates aren’t that informative). Of those aged 25+, consequentialists give more than non-consequentialists, p = .04, 1.7x the giving ($56 vs. $32), n = 526.
One more analysis, inspired in part by Gwern’s comment here.
In my first analysis I broke people into two groups using a cutoff at the bottom of the distribution, giving anything vs. giving nothing. But if we care about how much money charities receive, then we should care more about the top of the distribution, because that’s where most of the money is coming from (80-20 rule and all).
So, let’s focus on the top of the distribution by putting a cutoff up there. $1000 is an especially convenient Schelling point, since it is the 80-20 point among donors: 21.8% of those who gave to charity gave $1000+, and they accounted for 81.4% of the donations.
1164 people answered the question about moral views, and 10.8% of those reported giving $1000 or more to charity (this counts those who left the charity question blank in the under-$1000 group). Breaking that down by consequentialism:
12.5% of consequentialists gave $1000+
7.7% of non-consequentialists gave $1000+
p = .01.
Earlier, I found that consequentialists and non-consequentialists are equally likely to give to charity (vs. giving nothing). But here, we see that consequentialists are more likely to be big-money donors ($1000+).
Age, income, and religiosity are also significantly predictive of giving $1000+, and consequentialism remains a significant predictor (p=.002) after controlling for them.
Logging the total charity donations:
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.
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.)
Interestingly for the memeplex theory (which seems very plausible), Consequentialists are also more likely to indicate understanding of the Newcombs Problem (by choosing one box or two box over not sure or no answer) [chi2 = 16.1393, p = 0.001] but not more likely to actually one box over two box (chi2 = 4.1755, p = 0.243).