Lots of cultures want to have at least one boy child, and might keep having children until they get one / give up, so that might help explain why women have more siblings than men?
Good point! My ex is Vietnamese and while she loves her daughter she explicitly wants a son. My understanding is that male children act as a retirement plan.
I once dug into why the CDC report on sexual abuse had such off a deviance between yearly and lifetime abuse suffered by men. Another study shed some light: among adults that were definitely abused as children, men were something like 1/4 as likely as women to respond “yes” when asked if they were absurd as children.
My hypothesis was our culture has ready narratives of victimhood for women so women remember their abuse better. Men don’t have those narratives or at least don’t benefit from them (sympathy isn’t offered to men or is harmful) so men don’t frame their lives that way and therefore forget or at least don’t remember without being promoted with more than a label they don’t apply to themselves.
I might mis-phrase things and be accidentally offensive here. I apologise if that happens, I have tried hard to avoid it, but I know I do it sometimes, and ask your patience.
If your theory about narratives affecting how people remember/reconstruct their memories is consistent, I would naively expect those who affiliate more with said narratives to have a stronger effect.
I know you limited things to politically liberal/progressive, but did you happen to ask about how strongly the people identified with feminism or related ideas? I'd expect politically liberal/conservative to be good proxy for this, but I wonder if a more targeted examination might elicit more clarity there - if "degree of feminism" and the variables of interest are correlated it might imply something?
(Obviously feminism isn't all about "women are all victims", I'm using it as a crude proxy for said narratives)
I thought that as well; we only saw one of the graphs for conservatives.
One of the attractions of red-pill is that it creates a victimhood narrative for men. Those things are always tempting; the world sucks in general, I think.
A) Male data point. I was spanked as a kid, experienced really intense medical trauma (bleeding disorder Hemophilia, in hospitals A LOT, thousands of IV needles (only delivery mechanism of my medicine), being held down by nurses to get needles in to me).
I eventually got over the trauma. by my mid-20s I'm pretty sure I didn't consider my childhood traumatic. Maybe if you asked me when I as 15 or something. I have a pretty positive look back on my life, but this is PROBABLY a coping mechanism or a philosophical software package.
It seems odd to me that women were abused/mistreated more as children. I just performed the "as a child, did I get spanked ____" and there is some subtle determination that happens between Rarely and Sometimes for me.
B) I would love if you could somehow get a team of researchers to interview random people at banks and Walmart to take the study to figure out how biased the online-survey-taking-women-from-tiktok sample set actually is. e.g. my girlfriend has a trades job, is the opposite of "hyper online" and also skews heavily away from almost all of the data points I've read in your articles.
Hey man, going to the hospital because of spanking sounds objectively traumatic, regardless of how well you have overcome it since then. We should distinguish between subjective perceptions of trauma and actual objective indicators of trauma. Both are important obviously, but it is still important to distinguish.
Hey I just read this many months later, but I think that the medical trauma is distinct from the signing here. Even if a spanking caused bleeding, it could not have caused hemophilia!
Well I wasn’t actually making a gendered commentary. I noticed another interesting correlation which was that the belief in the supernatural decreases as class standing increases UNTIL we get to the elite class where it reverses and belief goes back up.
My hunch is that the upper class is mostly highly educated individuals but the elite often come from old money/are less skilled professionals. Like, physicists and geneticists and other highly educated jobs are not going to be in the elite class.
Sorry I didn’t mean to imply your were. Actually in retrospect, my ex is an elite and also a dick so gender does apply to my experience.
Yeah probably that is how the elite got there. The elites I know all had rich or powerful parents. They also hate themselves and vacílate from feeling superior to everyone to feeling everyone is better than them.
I have a hypothesis that heirs tend to be crappy people because while their grandparents were exceptionally competent, they’re not, so maintaining their wealth and competing with excellent people pushes them toward more brutal strategies.
Fwiw, my sisters both report being emotionally and verbally abused by my dad, while I do not. Neither my mom nor I remember any notable difference in how my dad treated me versus my sisters, and when I've gotten my sisters to talk about specific events they considered abusive (this was very hard, btw, they mostly wanted to just talk about how they "experienced abuse" in abstract terms), they reported events where I remember the event, sometimes as a fellow "victim" sometimes as an observer. We have more or less the same factual recall of the events. They just perceive them as "abuse" while I perceive them as "my dad being an asshole". So, yeah, the interpretation of my sisters wanting to/believing they fit into the victim/survivor/whatever narrative whereas men take a more decouple-y perspective definitely scans with my experience.
Also relevant, my sisters insist against my judgement to the contrary to my insistence that I also experienced abuse. Actually, this might make an interesting follow-up. Something like "did your [male/female] siblings experience [emotional, verbal, physical] abuse?" If the "experiences were identical, framing is different" hypothesis is true, then I would expect to see women reporting their male siblings being abused at similar rates to themselves being abused, and vice versa.
Possibly relevant demographic info: My siblings and I are aged 26-30, two sisters, I was the only boy and middle child. My sisters would probably describe as "very liberal" or maybe just "liberal" (they are very liberal, but are also in a closed bubble and I'm not sure if they'd correct for their peer group also being very liberal). I go back and forth between "somewhat conservative" and "no affiliation".
Since I am not part of your memories, I can't speak to them. But - knowing what I have been through with a covert abuser, I can understand why there might be a difference in how you see the same events as him being an asshole, and they recall abuse. Covert abusers know how to land abuse directly on target and not hit others nearby. I have a favorite comic song that has this line: "It's not the fall but landing that will alter social standing." You might see him fall, and he walks away from you seeming okay, just being him. But what he did, landed on them. Which altered his relationship to them. Not you. This happens all the time. Family members that are untouched by the landing cannot comprehend why someone else in the family is upset, because they can't get the point of what happened. But the victims will know, in context to other memories, more about what is going on. Why the landing is so hard. If they try to explain it? It will not work unless someone listening to them has been in the same position. It is a psychological abuse that is invisible to bystanders. But if you know what's going on, it can be devastating. The people most undermined in this way are probably usually women. It is baffling and humiliating to go through. It is difficult to get anyone else to believe.
Yeah, that is a good alternative explanation. Not really applicable in my case, since I also had fairly similar-seeming interactions with my dad where I was the target. Unless he was just worse at getting those hits to land on me? Which I suppose is possible. Laying gender aside, my sisters are generally more neurotic, worse at decoupling, etc. Not sure how that changes the accounting in the abuse discussion, since the concept is a bit under-specified. But, regardless, useful insight, thanks for sharing.
Yeah, I was thinking that too. I noticed that most people were never abused, though; it looks like there's a small fraction that's having all the bad times and they're more heavily female.
That could mean several different things. Do you mean the girls think they're personally poorer or that the family is poorer relative to other families? Are the girls getting less resources? Might boys be more optimistic while girls are more sensitive to the families' actual status? I'm probably missing some possibilities.
There are many, many possibilities - but the most probable possibility is that because the girls were higher in Emotionality, that made them more sensitive to the negative experiences they had.
Did I tell you about the time I was goofing off with one of my friends and that dude punched me in the face? I stared at him, and he backed down. Maybe I had a black eye? That's the end of the story. No therapy, not safe spaces, no bandages, no tearful diary entries. Then eventually I made, like, fifty pies out of crab apples I picked from the side of the road and people's yards without caring whether they were full of worms or anthrax or COVID. Dudes are like this.
First, a datapoint: I'm a male and i'm *very* nostalgic. I yearn for pretty much any period from more than 2 years ago (and it's unrelated to any particular event, it's a rolling 2 years period before I start having a rose tint when I look back at it).
Then a nitpick:
>This one is weird. We’re looking only at the most liberal, so it’s not like we’re getting conservative and discipline-heavy girls vs. freewheeling boys.
If we assume that conservative do most of the punishement, then it's possible that conservative do punish boys more than girls, and that the marginal punishment dished by liberals is somehow more targeted toward girls (and your demographic mostly catch these). I don't really believe it's possible, but it's maybe worth investigating.
Not an area I'm familiar with, but I did a quick scan of the literature and it appears conservative households are more likely to spank than liberal/progressive ones. This doesn't disprove your hypothesis, but since we know politics tends to be heritable, it seems unlikely?
One thing I've started focusing a lot on in surveys is the question of "what is it about?". For example, when children get spanked, it's for doing something they're not allowed to do, right? So I've come to believe it's very incomplete to just ask about the frequency and not what it was about.
I'd be curious about e.g. whether men and women remember being spanked for the same things.
What's your sample size for men and women respectively? If you have a selection bias with respect to both gender and some other variable, then this can lead to collider bias which can introduce artifactual correlations.
Collider bias is technically a very general bias which encompasses a huge variety of statistical paradoxes, but there's a special case that comes up in sampling. Let's take an easy to understand example.
Imagine you want to study rationalists and kink, so you post a survey to LessWrong and to FetLife. It seems reasonable to assume that use of LessWrong and FetLife would be uncorrelated or positively correlated, but if you actually did this, most likely they would be negatively correlated in your resulting dataset. Why?
It's due to the structure of the sampling. One way to get into the survey is through LessWrong; people who use LessWrong a lot are much more likely to get into it via this path, and therefore they will be overrepresented in the sample. Meanwhile, another way to get into the survey is through FetLife; people who use FetLife a lot are much more likely to get into it via this path, and therefore they too will be overrepresented in the sample. But this means that your overall sample actually consists of two non-overlapping subsamples, one of which uses LessWrong a lot and one of which uses FetLife a lot, and you would detect this sampling structure in a negative correlation between the two interests.
More generally, when you've got two variables A and B (e.g. LessWrong use and FetLife use) which both influence some variable C (such as taking your survey), and you "condition" on C ("conditioning" in the case of sampling means analyzing data from your survey, but in other contexts such as regression it might mean including the variable in the regression model), then this can induce weird relationships between A and B, I think usually a negative correlation but strictly speaking sometimes it can also induce other stuff.
Implication: clearly by the gender ratio of your survey, your survey filters for being female. If your survey also filters for being upper class, then you've got a collider: female -> taking your survey <- upper class. This would be expected to induce some artifactual correlations between being female and being upper class, most likely negative ones, i.e. if you expect your survey disproportionately has upper-class people then you should also expect that your data overstates the correlation between being lower class and being female.
Usually the distortion due to collider bias isn't huge, but the effects you are working with in this post aren't huge either, so maybe the collider is strong enough to explain it. Sometimes I think surveys should have in-depth questions on how one encountered the survey to help control for collider bias, but in practice I haven't really done so and I am not sure it is feasible.
Another good example of collider bias comes up in mortality. If you do something to prevent one disease, then this is going to increase the likelihood that your death will be caused by another disease. Why? Because death is a collider, so when we condition on death, a decrease in one cause of death means an increase in other causes of death. In order to solve this, epidemiologists look at death rate per year for each disease rather than cause of death, as that avoids conditioning on death.
Another name for collider bias is Berkson's paradox, if you want to read more. I also wrote an article on it in a different context which might help:
awesome, thank you. I've read about this bias before and could probably convincingly pretend to understand it, but I still don't *deeply* understand the way this applies to non-obvious examples (like class and gender). I'll read the linked post!
I think my linked post contains a diagram that is pretty helpful for understanding it. You can also do a google image search for "Berkson's paradox" to see a bunch of more examples.
Also, a long-term goal you should consider is learning how to write simulation studies. Basically, by writing a computer program which simulates how you think the dynamics play out, you can empirically test what biases this would induce in your data. I've just quickly written up a simulation study for the class/gender data which you maybe can use, though I haven't tuned the parameters to be particularly realistic for your case:
import numpy as np
import matplotlib.pyplot as plt
# constants for configuring the simulation
male = 0; female = 1
N = 4000
# generate data for simulated people
gender = np.random.binomial(1, 0.5, N)
SES = np.random.normal(1, 1.5, N)
# since the code generates them independently, this corresponds to an assumption that they are independent in the general population
# people randomly join your survey based on characteristics
I read your Lesswrong article. I think you are right about that bias. That it is a paradox. One we, as humans, will always have difficulty with. We might individually have similar ways of processing information and any resulting emotions about that information in our minds, but none of us can have identical ones. Which is why, personally, I can't take studies like this seriously. Maybe as indicators of possible behavior, but they don't prove their cases.
I’m reminded of a theory in evolutionary biology asserting that in times of plenty, people will have more men, because the upside of a really great man is a very large number of grandchildren, and in times of squalor people will have more women, because there’s a limit to how reproductively unproductive women could be.
I think it's misleading to restrict this to liberals and call that "controlling for political orientation". When you include everyone I think the spanking trend reverses, and men are spanked more.
I'm sorry I do not have the study on hand, so please just count my comment as something to consider rather than evidence. I have read a study which showed that women do grow up in poorer households than men. This is because when a man has children, his salary goes up as a result of him working harder to support the household. However, his salary goes up significantly more if he has a son than if he has a daughter. I have also seen research showing that parents spend more time with sons than daughters, supporting the neglect results.
This is silly. The survey asks for if men and women have such and such experiences in their childhood, women remember their childhood differently than men, so the survey concludes people are wrong about their childhood? Wrong; not every family has more than one child and they could come from households that do or do not have people of the other gender
To me, the results seem to make very good sense; men and women everywhere are influenced by the world around them regardless of many other factors. If a family only comtains females or contains more females, it is more likely that family will be lower class because of the pervasiveness of gender inequity.
I realize it tries to control for some things, like parents reporting their boys being spanked more often by their parents and being more rambunctious, but it fails to consider that spanking does not always occur in the home (Catholic schools, for instance. It doesn't fully control for the parents beliefs or behaviors or where they are IN these countries even when we know they are white and in a wealthy country) and just because the parents say they spank the boy more, that doesn't mean that they do, just that they associate boys with being boys and probably assume in their head that the boy gets punished more when in reality, they let them off the hook more often than not.
Lots of cultures want to have at least one boy child, and might keep having children until they get one / give up, so that might help explain why women have more siblings than men?
Good point! My ex is Vietnamese and while she loves her daughter she explicitly wants a son. My understanding is that male children act as a retirement plan.
I once dug into why the CDC report on sexual abuse had such off a deviance between yearly and lifetime abuse suffered by men. Another study shed some light: among adults that were definitely abused as children, men were something like 1/4 as likely as women to respond “yes” when asked if they were absurd as children.
My hypothesis was our culture has ready narratives of victimhood for women so women remember their abuse better. Men don’t have those narratives or at least don’t benefit from them (sympathy isn’t offered to men or is harmful) so men don’t frame their lives that way and therefore forget or at least don’t remember without being promoted with more than a label they don’t apply to themselves.
Neuroticism is a better explanation.
I might mis-phrase things and be accidentally offensive here. I apologise if that happens, I have tried hard to avoid it, but I know I do it sometimes, and ask your patience.
If your theory about narratives affecting how people remember/reconstruct their memories is consistent, I would naively expect those who affiliate more with said narratives to have a stronger effect.
I know you limited things to politically liberal/progressive, but did you happen to ask about how strongly the people identified with feminism or related ideas? I'd expect politically liberal/conservative to be good proxy for this, but I wonder if a more targeted examination might elicit more clarity there - if "degree of feminism" and the variables of interest are correlated it might imply something?
(Obviously feminism isn't all about "women are all victims", I'm using it as a crude proxy for said narratives)
I thought that as well; we only saw one of the graphs for conservatives.
One of the attractions of red-pill is that it creates a victimhood narrative for men. Those things are always tempting; the world sucks in general, I think.
A) Male data point. I was spanked as a kid, experienced really intense medical trauma (bleeding disorder Hemophilia, in hospitals A LOT, thousands of IV needles (only delivery mechanism of my medicine), being held down by nurses to get needles in to me).
I eventually got over the trauma. by my mid-20s I'm pretty sure I didn't consider my childhood traumatic. Maybe if you asked me when I as 15 or something. I have a pretty positive look back on my life, but this is PROBABLY a coping mechanism or a philosophical software package.
It seems odd to me that women were abused/mistreated more as children. I just performed the "as a child, did I get spanked ____" and there is some subtle determination that happens between Rarely and Sometimes for me.
B) I would love if you could somehow get a team of researchers to interview random people at banks and Walmart to take the study to figure out how biased the online-survey-taking-women-from-tiktok sample set actually is. e.g. my girlfriend has a trades job, is the opposite of "hyper online" and also skews heavily away from almost all of the data points I've read in your articles.
Hey man, going to the hospital because of spanking sounds objectively traumatic, regardless of how well you have overcome it since then. We should distinguish between subjective perceptions of trauma and actual objective indicators of trauma. Both are important obviously, but it is still important to distinguish.
Hey I just read this many months later, but I think that the medical trauma is distinct from the signing here. Even if a spanking caused bleeding, it could not have caused hemophilia!
The elite are looking... self-absorbed and narcissistic here 👀
Elite men being stupid pricks relative to upper class... yep that’s been my experience.
Well I wasn’t actually making a gendered commentary. I noticed another interesting correlation which was that the belief in the supernatural decreases as class standing increases UNTIL we get to the elite class where it reverses and belief goes back up.
My hunch is that the upper class is mostly highly educated individuals but the elite often come from old money/are less skilled professionals. Like, physicists and geneticists and other highly educated jobs are not going to be in the elite class.
I could be wrong though
Sorry I didn’t mean to imply your were. Actually in retrospect, my ex is an elite and also a dick so gender does apply to my experience.
Yeah probably that is how the elite got there. The elites I know all had rich or powerful parents. They also hate themselves and vacílate from feeling superior to everyone to feeling everyone is better than them.
I have a hypothesis that heirs tend to be crappy people because while their grandparents were exceptionally competent, they’re not, so maintaining their wealth and competing with excellent people pushes them toward more brutal strategies.
To be fair though, the chart about caring less about others does show elite men to be worse than women 🤷♂️
Yea that checks out. No worries!
Fwiw, my sisters both report being emotionally and verbally abused by my dad, while I do not. Neither my mom nor I remember any notable difference in how my dad treated me versus my sisters, and when I've gotten my sisters to talk about specific events they considered abusive (this was very hard, btw, they mostly wanted to just talk about how they "experienced abuse" in abstract terms), they reported events where I remember the event, sometimes as a fellow "victim" sometimes as an observer. We have more or less the same factual recall of the events. They just perceive them as "abuse" while I perceive them as "my dad being an asshole". So, yeah, the interpretation of my sisters wanting to/believing they fit into the victim/survivor/whatever narrative whereas men take a more decouple-y perspective definitely scans with my experience.
Also relevant, my sisters insist against my judgement to the contrary to my insistence that I also experienced abuse. Actually, this might make an interesting follow-up. Something like "did your [male/female] siblings experience [emotional, verbal, physical] abuse?" If the "experiences were identical, framing is different" hypothesis is true, then I would expect to see women reporting their male siblings being abused at similar rates to themselves being abused, and vice versa.
Possibly relevant demographic info: My siblings and I are aged 26-30, two sisters, I was the only boy and middle child. My sisters would probably describe as "very liberal" or maybe just "liberal" (they are very liberal, but are also in a closed bubble and I'm not sure if they'd correct for their peer group also being very liberal). I go back and forth between "somewhat conservative" and "no affiliation".
Since I am not part of your memories, I can't speak to them. But - knowing what I have been through with a covert abuser, I can understand why there might be a difference in how you see the same events as him being an asshole, and they recall abuse. Covert abusers know how to land abuse directly on target and not hit others nearby. I have a favorite comic song that has this line: "It's not the fall but landing that will alter social standing." You might see him fall, and he walks away from you seeming okay, just being him. But what he did, landed on them. Which altered his relationship to them. Not you. This happens all the time. Family members that are untouched by the landing cannot comprehend why someone else in the family is upset, because they can't get the point of what happened. But the victims will know, in context to other memories, more about what is going on. Why the landing is so hard. If they try to explain it? It will not work unless someone listening to them has been in the same position. It is a psychological abuse that is invisible to bystanders. But if you know what's going on, it can be devastating. The people most undermined in this way are probably usually women. It is baffling and humiliating to go through. It is difficult to get anyone else to believe.
Yeah, that is a good alternative explanation. Not really applicable in my case, since I also had fairly similar-seeming interactions with my dad where I was the target. Unless he was just worse at getting those hits to land on me? Which I suppose is possible. Laying gender aside, my sisters are generally more neurotic, worse at decoupling, etc. Not sure how that changes the accounting in the abuse discussion, since the concept is a bit under-specified. But, regardless, useful insight, thanks for sharing.
Maybe I missed something, but why is the idea that girls might be treated worse than boys on the average not plausible?
I think it was, but the social class and the spanking question throws the rest into doubt.
Yeah, I was thinking that too. I noticed that most people were never abused, though; it looks like there's a small fraction that's having all the bad times and they're more heavily female.
One issue is that girls think they’re poorer than boys even when they came from the same household.
That could mean several different things. Do you mean the girls think they're personally poorer or that the family is poorer relative to other families? Are the girls getting less resources? Might boys be more optimistic while girls are more sensitive to the families' actual status? I'm probably missing some possibilities.
There are many, many possibilities - but the most probable possibility is that because the girls were higher in Emotionality, that made them more sensitive to the negative experiences they had.
Did I tell you about the time I was goofing off with one of my friends and that dude punched me in the face? I stared at him, and he backed down. Maybe I had a black eye? That's the end of the story. No therapy, not safe spaces, no bandages, no tearful diary entries. Then eventually I made, like, fifty pies out of crab apples I picked from the side of the road and people's yards without caring whether they were full of worms or anthrax or COVID. Dudes are like this.
First, a datapoint: I'm a male and i'm *very* nostalgic. I yearn for pretty much any period from more than 2 years ago (and it's unrelated to any particular event, it's a rolling 2 years period before I start having a rose tint when I look back at it).
Then a nitpick:
>This one is weird. We’re looking only at the most liberal, so it’s not like we’re getting conservative and discipline-heavy girls vs. freewheeling boys.
If we assume that conservative do most of the punishement, then it's possible that conservative do punish boys more than girls, and that the marginal punishment dished by liberals is somehow more targeted toward girls (and your demographic mostly catch these). I don't really believe it's possible, but it's maybe worth investigating.
Or maybe being punished as a child skews the adult in a liberal direction?
Not an area I'm familiar with, but I did a quick scan of the literature and it appears conservative households are more likely to spank than liberal/progressive ones. This doesn't disprove your hypothesis, but since we know politics tends to be heritable, it seems unlikely?
One thing I've started focusing a lot on in surveys is the question of "what is it about?". For example, when children get spanked, it's for doing something they're not allowed to do, right? So I've come to believe it's very incomplete to just ask about the frequency and not what it was about.
I'd be curious about e.g. whether men and women remember being spanked for the same things.
What's your sample size for men and women respectively? If you have a selection bias with respect to both gender and some other variable, then this can lead to collider bias which can introduce artifactual correlations.
It was 43k and 138k. I don't think I understand collider bias but would like to learn
Collider bias is technically a very general bias which encompasses a huge variety of statistical paradoxes, but there's a special case that comes up in sampling. Let's take an easy to understand example.
Imagine you want to study rationalists and kink, so you post a survey to LessWrong and to FetLife. It seems reasonable to assume that use of LessWrong and FetLife would be uncorrelated or positively correlated, but if you actually did this, most likely they would be negatively correlated in your resulting dataset. Why?
It's due to the structure of the sampling. One way to get into the survey is through LessWrong; people who use LessWrong a lot are much more likely to get into it via this path, and therefore they will be overrepresented in the sample. Meanwhile, another way to get into the survey is through FetLife; people who use FetLife a lot are much more likely to get into it via this path, and therefore they too will be overrepresented in the sample. But this means that your overall sample actually consists of two non-overlapping subsamples, one of which uses LessWrong a lot and one of which uses FetLife a lot, and you would detect this sampling structure in a negative correlation between the two interests.
More generally, when you've got two variables A and B (e.g. LessWrong use and FetLife use) which both influence some variable C (such as taking your survey), and you "condition" on C ("conditioning" in the case of sampling means analyzing data from your survey, but in other contexts such as regression it might mean including the variable in the regression model), then this can induce weird relationships between A and B, I think usually a negative correlation but strictly speaking sometimes it can also induce other stuff.
Implication: clearly by the gender ratio of your survey, your survey filters for being female. If your survey also filters for being upper class, then you've got a collider: female -> taking your survey <- upper class. This would be expected to induce some artifactual correlations between being female and being upper class, most likely negative ones, i.e. if you expect your survey disproportionately has upper-class people then you should also expect that your data overstates the correlation between being lower class and being female.
Usually the distortion due to collider bias isn't huge, but the effects you are working with in this post aren't huge either, so maybe the collider is strong enough to explain it. Sometimes I think surveys should have in-depth questions on how one encountered the survey to help control for collider bias, but in practice I haven't really done so and I am not sure it is feasible.
Another good example of collider bias comes up in mortality. If you do something to prevent one disease, then this is going to increase the likelihood that your death will be caused by another disease. Why? Because death is a collider, so when we condition on death, a decrease in one cause of death means an increase in other causes of death. In order to solve this, epidemiologists look at death rate per year for each disease rather than cause of death, as that avoids conditioning on death.
Another name for collider bias is Berkson's paradox, if you want to read more. I also wrote an article on it in a different context which might help:
https://www.lesswrong.com/posts/wM4bcDxEh75NDkhjo/are-smart-people-s-personal-experiences-biased-against
awesome, thank you. I've read about this bias before and could probably convincingly pretend to understand it, but I still don't *deeply* understand the way this applies to non-obvious examples (like class and gender). I'll read the linked post!
I think my linked post contains a diagram that is pretty helpful for understanding it. You can also do a google image search for "Berkson's paradox" to see a bunch of more examples.
Also, a long-term goal you should consider is learning how to write simulation studies. Basically, by writing a computer program which simulates how you think the dynamics play out, you can empirically test what biases this would induce in your data. I've just quickly written up a simulation study for the class/gender data which you maybe can use, though I haven't tuned the parameters to be particularly realistic for your case:
import numpy as np
import matplotlib.pyplot as plt
# constants for configuring the simulation
male = 0; female = 1
N = 4000
# generate data for simulated people
gender = np.random.binomial(1, 0.5, N)
SES = np.random.normal(1, 1.5, N)
# since the code generates them independently, this corresponds to an assumption that they are independent in the general population
# people randomly join your survey based on characteristics
selection_probability = 1/(1+np.exp(-(gender+SES-2)))
surveyed = np.random.binomial(1, selection_probability)
plt.subplot(131)
plt.scatter(gender[surveyed == 0], SES[surveyed == 0], label="people not included in your survey", alpha=0.05)
offset = 0.05 # makes it easier to see the distribution
plt.scatter(gender[surveyed == 1]+offset, SES[surveyed == 1], label="people included in your survey", alpha=0.05)
plt.xticks([male, female], ["male", "female"])
plt.ylabel("childhood socioeconomic status")
plt.legend()
plt.subplot(132)
plt.bar([-0.1, 0.9], [np.mean(SES[gender == male]), np.mean(SES[gender == female])], 0.2, label="all people")
plt.bar([0.1, 1.1], [np.mean(SES[(gender == male) & (surveyed == 1)]), np.mean(SES[(gender == female) & (surveyed == 1)])], 0.2, label="people included in your survey")
plt.xticks([male, female], ["male", "female"])
plt.ylabel("childhood socioeconomic status")
plt.legend()
plt.subplot(133)
plt.scatter(SES[gender == male], selection_probability[gender==male], label="men")
plt.scatter(SES[gender == female], selection_probability[gender==female], label="women")
plt.xlabel("childhood socioeconomic status")
plt.ylabel("probability of taking survey")
plt.legend()
plt.show()
I read your Lesswrong article. I think you are right about that bias. That it is a paradox. One we, as humans, will always have difficulty with. We might individually have similar ways of processing information and any resulting emotions about that information in our minds, but none of us can have identical ones. Which is why, personally, I can't take studies like this seriously. Maybe as indicators of possible behavior, but they don't prove their cases.
Thanks for talking about this - I never knew what it was called. The image at your link is perfect.
Potentially false memories
It's easy to look backwards and apply today's feelings and insecurities to the past
And to ignore our own negatives
" I cropped the sample down to people aged 19-26" Oh, no wonder the results seem so distant from reality.
I’m reminded of a theory in evolutionary biology asserting that in times of plenty, people will have more men, because the upside of a really great man is a very large number of grandchildren, and in times of squalor people will have more women, because there’s a limit to how reproductively unproductive women could be.
I think it's misleading to restrict this to liberals and call that "controlling for political orientation". When you include everyone I think the spanking trend reverses, and men are spanked more.
I didn't notice that - you're right, there are far better ways of controlling for political orientation.
I'm sorry I do not have the study on hand, so please just count my comment as something to consider rather than evidence. I have read a study which showed that women do grow up in poorer households than men. This is because when a man has children, his salary goes up as a result of him working harder to support the household. However, his salary goes up significantly more if he has a son than if he has a daughter. I have also seen research showing that parents spend more time with sons than daughters, supporting the neglect results.
This is silly. The survey asks for if men and women have such and such experiences in their childhood, women remember their childhood differently than men, so the survey concludes people are wrong about their childhood? Wrong; not every family has more than one child and they could come from households that do or do not have people of the other gender
To me, the results seem to make very good sense; men and women everywhere are influenced by the world around them regardless of many other factors. If a family only comtains females or contains more females, it is more likely that family will be lower class because of the pervasiveness of gender inequity.
I realize it tries to control for some things, like parents reporting their boys being spanked more often by their parents and being more rambunctious, but it fails to consider that spanking does not always occur in the home (Catholic schools, for instance. It doesn't fully control for the parents beliefs or behaviors or where they are IN these countries even when we know they are white and in a wealthy country) and just because the parents say they spank the boy more, that doesn't mean that they do, just that they associate boys with being boys and probably assume in their head that the boy gets punished more when in reality, they let them off the hook more often than not.