News for hackers. Stuff that matters.
Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @01:28AM (#9986071)
In every work environment which has an clear gender imbalance we see the same stories of the minority group being pushed out by the majority group.

This isn't true at all. Lots of fields had clear gender imbalances and explicit hostility, and yet the minority gender had little trouble entering. For example, medicine, law, pharmacy, HR, advertising, professor of subjects which don't use math, etc.

Re:Django Girls one year later by keypusher on Saturday August 01, 2015 @02:07AM (#9986127)
Are you sure about that? I think the difference is that the pioneering women who opened the doors of fields such as medicine and law did so over 100 years ago. The minority gender had lots of trouble entering, and up until recent times suffered from many of the same things that women complain about in the programming community.
Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @03:13AM (#9986236)
Software is well past the point that medicine was at in 1915. In fact, computing specifically had lots of women in it's early days - probably more than medicine had at the same time.

Any theory which predicts that math/physics/cs/EE has few women is inherently flawed if it doesn't also predict that medicine/law/advertising have plenty of women.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @03:28AM (#9986261)
> Any theory which predicts that math/physics/cs/EE has few women is inherently flawed if it doesn't also predict that medicine/law/advertising have plenty of women.

I am not sure where you're going with this, but you can't make such a comparison at a given moment in time. The reason is that struggles usually follow power, and so it is likely that at any point in time, a marginalized group would concentrate its efforts to penetrate positions that imbue power. As the power structure shifts, so does the struggle. You can't equate professions that hold different amounts of power at different times. Certainly up until some decades ago, law and medicine carried much more prestige and influence -- and hence, power -- than math and the physical sciences. In fact, software's gain of prestige correlates quite well with the drop in women participation. It is quite anomalous in that the existence of the field long predated its power. Law and medicine, OTOH, are professions that have been seats of power for almost a millennium.

A model that correlates low female participation with high power -- and later, increased participation after a struggle -- predicts both observations. It is not such a surprising or sophisticated theory; historians have seen this pattern so often that it is almost banal. This pattern is so regular, that it's relatively easy for us to predict the future: fifty years from now, assuming the power wielded by software remains high, women participation will equal men's, and it will be hard for people to believe that that hasn't always been the case. Unfortunately, even though the result of the fight is certain, we still need to fight it. That is the price history exacts from us.

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @03:41AM (#9986285)
Can you state your theory clearly enough that I can directly apply it? I have no idea what you are trying to say. Make sure to include a clear definition of "power", and make sure this definition is clear enough that I can at the very least evaluate "more" and "less" power.

(I.e., I have no idea why you consider academic physics to have more power than academic biology.)

I also have no idea what you mean by "a marginalized group would concentrate its efforts". Are you implying that women don't make individual career choices based on their interest and aptitude, but instead "concentrate [their] efforts" in some sort of collectivist power grab?

That's a bit scary. I always thought women were just people like me, making selfish individual decisions to maximize their money/lifestyle/career enjoyment.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @06:21AM (#9986540)
> Can you state your theory clearly enough that I can directly apply it?

Certainly. Non-hegemonic groups are marginalized from seats of power. After a long struggle, they may increase participation. Put another way: groups with more power tend to preserve and increase the power difference between them and less powerful groups, while the less-powerful groups may gain a share of the power after a struggle.

Sadly, it's not my theory, but one of the most elementary theories -- backed by countless evidence -- of the social sciences, supported by studies in anthropology, sociology and history (with some backup from social psychology, too).

Can you state your theory clearly, or, at the very least, state what evidence compels you to doubt the current scientific consensus?

> Make sure to include a clear definition of "power", and make sure this definition is clear enough that I can at the very least evaluate "more" and "less" power.

I have done so on numerous occasions in the past in our conversations. Please refer to them or look up "power" on Wikipedia. The short description is influence, and if you'd like a description with a more quantitative "feel", I'd say the power a person has is the number of people they can influence indirectly (i.e. graph reach) summed over the magnitude in the change of behavior they cause in each affected individual.

> I.e., I have no idea why you consider academic physics to have more power than academic biology

I don't. See my other response to you.

> Are you implying that women don't make individual career choices based on their interest and aptitude, but instead "concentrate [their] efforts" in some sort of collectivist power grab?

I'm stating the much-observed, well-known fact, that society pushes and directs us in various directions. What we see as "free choice" is, in fact, "free choice under societal pressures and restrictions". That behavior is not collective but individual, only biased. Just like any fair coin is free to make a choice how to land, yet we can make very accurate predictions on the result of a thousand coin tosses. Collective behavior is not always the result of collective decisions.

> I always thought women were just people like me, making selfish individual decisions to maximize their money/lifestyle/career enjoyment.

All of us make selfish individual decisions, but the (probabilistic) fitness function is largely determined by large-scale, emergent properties of the system. Think of individual particle movement and temperature or of Brownian motion. There are no collective decisions involved -- only individual "decisions" and local interaction -- yet the end result is that global factors heavily influence the distribution in behavior of the individuals. The effect of social influence on personal choice has been heavily studied in social psychology, and, in fact, in zoology too.

Re:Django Girls one year later by sfk on Saturday August 01, 2015 @07:48AM (#9986708)
Seats of power? Any woman can start and maintain an OSS project like xterm. They don't, because it's a thankless task and either they aren't interested enough or they figure out the "thankless" part faster than men.
Re:Django Girls one year later by pbhjpbhj on Saturday August 01, 2015 @08:34AM (#9986808)
>or they figure out the "thankless" part faster than men //

Or men figure out the "thankless" part equally quickly but aren't as bothered to receive encouragement from others as the women?

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @09:03AM (#9986887)
You've only provided one clear definition of "power", namely "influences the behavior of others". This is simply a relation. It is not an ordering or a cardinality. Yet you then turn around and use "power" in an ordinal way ("power difference") and sometimes even a cardinal way.

The definition you give here almost does that, but how do I assign a number to "magnitude in the change of behavior"? You've not defined a "hegemonic group" either.

Anyway, since I can't quite understand your theory, maybe you can clearly state a hypothetical observation which would prove your theory incorrect. That's usually a good way to illustrate what a theory does and does not predict.

For example, to disprove my "more math less women" theory, you should find a broad category of similar careers, run linear regression on math content vs % female, and find a flat or upward sloping line. I.e., the same analysis as was done on Scott Alexander's post (see my link), but with the opposite result.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @09:49AM (#9987011)
> but how do I assign a number to "magnitude in the change of behavior"?

You don't. Read what I wrote.

> You've not defined a "hegemonic group" either.

If you're interested, look it up. I feel a strong sense of deja-vu.

> maybe you can clearly state a hypothetical observation which would prove your theory incorrect

I think I've done it twice in the past when talking to you, so I'll only provide a short abstract here (and again, it's not my theory but the consensus theory): If groups with power were not to resist social mobility (i.e. for less powerful groups to obtain more power) we'd see a constant shift in power distribution, which would be very fluid, while what we observe is the complete opposite: a very stable power structure that then breaks and gets restructured relatively very rapidly during revolutions that are rare. So the pattern is long periods of stability and then quick changes brought about by some explosive event or a demonstrable pattern of slow power acquisition through many struggles. When it comes to women, the pattern is often the latter (there's a simple explanation for that), but you can see how women gain access to politics, law, medicine etc. only after very long fights against very strong opposition (that invariably cite innate ability as the barrier).

> For example, to disprove my "more math less women" theory, you should find a broad category of similar careers, run linear regression on math content vs % female, and find a flat or upward sloping line. I.e., the same analysis as was done on Scott Alexander's post (see my link), but with the opposite result.

Two problems with that. First, yours is not a theory but a description. Second, it is not a very relevant one because it doesn't explain the constant change in participation, nor the gender-gap in the software industry. Finally, while "more math less women" may have always held on average, the slope has changed. I don't think anybody minds if the gender distribution in physics is 60-40%, even if that difference is entirely due to innate ability. So your description does not even compete with mine because it doesn't even try to explain the same phenomenon, which is the one we care the most to explain.

That has little to do with the 4:1 or even 10:1 difference we see in the software industry.

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @10:15AM (#9987086)
You don't.

Then we are back to power being a non-ordinal concept.

...more power...

If power is not an ordinal concept, this phrase is meaningless.

I'm also confused as to how you can even determine that such a "constant shift" isn't happening - in terms of edges in your influence graph, they shift regularly. Yesterday I met a new person at work and we now interact and influence each other. The power digraph has gained 2 edges. Could you explain how you quantify this?

Re:Django Girls one year later by pron on Saturday August 01, 2015 @11:18AM (#9987316)
> If power is not an ordinal concept, this phrase is meaningless.

You're reading things that I'm not saying. It is an ordinal concept, but we don't have the ability to measure it well, only notice large differences. You can think of it as a quantity for which we have very inaccurate measurement devices.

> I'm also confused as to how you can even determine that such a "constant shift" isn't happening - in terms of edges in your influence graph, they shift regularly. Yesterday I met a new person at work and we now interact and influence each other. The power digraph has gained 2 edges. Could you explain how you quantify this?

Because in this discussion we are talking about groups (social mobility in individuals is an interesting, yet somewhat different discussion), and when we look at a large number of people, it's easier to approximate power through proxies such as income, percentage of powerful roles such as CEOs, leading journalists etc. (it's harder to compare power in individuals, because it's hard to tell which is more influential, a media personality or the CEO of a large company, especially as the latter's influence might sometimes be hidden).

Shifts in power from nobility to the bourgeois class, or in the relative cultural and technological power of Europe, were sudden changes of relative homeostases. Similarly, you can track power distribution between men and women, or between whites and blacks in the US. They point at very stable states that are violently changed, or (usually in the case of women), a faster rate of smaller revolutions (or a more steady rate of change). The difference between male/female and race hegemonies is also well understood (and seen in many other cultures), because the classification of women as "strangers" is always very different from that of actual foreign ethnicities.

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @11:44AM (#9987429)
Previously you said this: If groups with power were not to resist social mobility (i.e. for less powerful groups to obtain more power) we'd see a constant shift in power distribution...

You now say this: "it's easier to approximate power through proxies such as income, percentage of powerful roles"

Lets work with this specific prediction. If a group has a tendency to enter a powerful role with probability p, and there are N members of this group, then the shifts in power distribution will have standard deviation (in percentage terms) of 1/sqrt(Np(1-p)).

That doesn't seem to agree with your "constant shift in power distribution" at all.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @02:34PM (#9988139)
Read it again. I said that if the consensus theory is wrong then we'd see constant power shifts. Instead, we see long stable periods disrupted by short rare, revolutions or a relatively slow, very visible struggle. Hence, hegemonies do resist social mobility. IIRC, you made the exact same mistake last time we discussed the issue.

Also, I fail to see how your formula is meaningful at all, as it has nothing to do with power dynamics. At best, it describes the distribution of the number of powerful people in a random sample. Well, OK, how is that helpful? It cannot describe any shifts in power as it assumes independence, where our understanding of power is anything but (you are more likely to be powerful if you're close to powerful people). The actual dynamics is a collection of non-linear processes in a complex system. A very, very crude example of a somewhat similar process is that of segregation[1] with the addition of random populations during the run. It is processes with strong correlations between neighboring agents that tend to show stability pockmarked with rare bifurcations, which is exactly what we see in human society.

When I went to study history in graduate school, I dreamt of applying math to historical events. Playing with cellular automata to recreate certain shifts in history is fun, but eventually teaches us little, as we cannot say much more about the model than the same qualitative descriptions historians give anyway. It is sometimes cool to figure out the values of certain variables in the model and assign them meaning (like the "despair level prior to the French Revolution", but even these numbers don't tell us much because the models are so sensitive we can't be sure we got them right, and even if we did, it's virtually impossible to measure the values directly rather than figuring them out retrospectively. Therefore, speaking about events in broad qualitative terms is often the best we can do (compound cellular automata's Turing completeness with the non-linear ODEs defining the automaton's rules and the difficulty measuring variables directly and you get very little predictive ability).

[1]: http://ccl.northwestern.edu/netlogo/models/Segregation

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @03:56PM (#9988455)
I said that if the consensus theory is wrong then we'd see constant power shifts.

This is clearly wrong - I demonstrated a stupidly simple alternate theory which disagrees with the consensus, but still lacks constant power shifts.

Rapid shifts in the composition of various fields would debunk basically every theory that presupposes human nature doesn't rapidly shift. As such, your theory proves very little, and other theories (e.g. mine) make the same predictions. So why cling to that theory, rather than, say, my theory of p=A - B x math_content + second_order_corrections?

My theory predicts more things (e.g., it predicts % women in academia, % women in finance, % women in tech vs non-tech roles) and is vastly simpler. How is that not a win?

Re:Django Girls one year later by pron on Saturday August 01, 2015 @11:17PM (#9989606)
> This is clearly wrong - I demonstrated a stupidly simple alternate theory which disagrees with the consensus, but still lacks constant power shifts.

Your theory didn't even explain the data (in the software industry), and had nothing to say about power. Basically, it can comfortably exist beside the consensus theory as a second-order effect (and the data actually looks like that is probably the case, explaining representation differences of up to 2:1).

> So why cling to that theory, rather than, say, my theory of p=A - B x math_content + second_order_corrections?

Two reasons: 1. your theory doesn't explain the anomalous software industry (neither the very low participation nor the sudden change) while the consensus theory does, and 2. your theory doesn't explain the lack of representation (and possible correction after well-documented struggles) in fields of high prestige and little math (such as law, medicine and politics). In all those fields: law, medicine, politics and software, women participation was (or is) far lower than 30%. It was/is lower than 20%.

You might think that singling out software is uninteresting from a statistical perspective, as in Scott Alexander's regression it was just a single (outlier) data point, but in fact, the software industry is larger than all "heavy math content" professions combined, and holds more power in today's society.

> My theory predicts more things (e.g., it predicts % women in academia, % women in finance, % women in tech vs non-tech roles) and is vastly simpler. How is that not a win?

Because it predicts fewer things, and not even what you claim it does. At best, it predicts fewer women than men in certain subjects, but that's it. However, we see that while the binary relation "fewer" may remain, how much fewer decreases over time. Second, it only holds (even as a binary prediction) when you look at the past few decades -- far too short to look at social change -- which means it does not explain the exclusion of women from law, medicine and politics. Third, it does not explain the way-too-low representation in CS, nor the drop in participation. The consensus theory explains all of it, for any time duration. It even explains interesting shifts we see in, say, education (more women as prestige drops, while in CS fewer women as prestige rises). Yes, it's more complicated, but fits reality better, plus it's supported by lots and lots of data.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @03:10AM (#9986233)
That is just completely false.

1. Women had enormous trouble entering those fields (and, in fact, are almost as well represented in the physical sciences). They won their position after great struggles. You'll note that some of those fields where women have had better success (after a long battle) have high governmental regulation and/or high rate of government employment, which makes it easier to allow women in through regulation after a successful political battle. The private sector has always been more conservative and more resistant to social change.

See: https://en.wikipedia.org/wiki/Women_in_medicine, https://en.wikipedia.org/wiki/Women_in_the_United_States_jud...

2. Women did not use to be so under-represented in software. There's a clear trend of women leaving software, starting in the mid-eighties: http://www.npr.org/sections/money/2014/10/21/357629765/when-... Women participation in software is on the decline, unlike any other white-collar profession, requiring math or not.

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @03:29AM (#9986263)
What, precisely, is completely false? According to you, women were heavily underrepresented in medicine/law/HR/advertising, but were NOT so underrepresented in computing.

According to belorn, the minority gender will be pushed out by the majority group. Based on what you say above, this effect should be larger in medicine/law/HR/advertising than in software. Hence, belorn's theory fails to predict reality.

Your theory of government employment fails to predict reality as well. Academic sciences also have large gender disparities; physics has very few women, while biology has lots. These gaps pretty clearly mirror the CS/medicine gaps outside academia. Yet government involvement in physics academia is pretty similar to government involvement in bio academia.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @03:41AM (#9986287)
The observation precisely follows the flow of power. See my other comment.

> Academic sciences also have large gender disparities; physics has very few women, while biology has lots.

Academia is less interesting to begin with because there is little difference in the power held by different fields (certainly between physics and biology today). Participation can therefore depend on many factors (perhaps side-products of some other process) that people have little interest to examine or change. Those differences are no different from the high participation of women relative to men in ballet dancing. It's hard to pinpoint the causes, and there's little reason or urgency to study them because there's little power imbalance involved (there is more funding directed to studying more serious diseases than less serious ones, unless there's some other good reason to study the less harmful ones).

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @03:51AM (#9986305)
There is plenty of reason to study them; they provide validation/rejection of our theories.

If you claim that "power" drives everything, yet fields with no power difference have disparities which directly mirror the disparities you claim are driven by "power", that suggests there is an underlying theory that doesn't use power at all and which predicts more things.

By Occams razor, the latter theory is the one which should be used.

Re:Django Girls one year later by Trombone12 on Saturday August 01, 2015 @06:35AM (#9986566)
Dude, you can't use Occam’s razor to select between one theory and "the idea of there being a better theory" o_0

Fine, there might be an underlying theory that explains everything better than the "power" theory, BUT YOU HAVE TO PROVIDE IT before you can argue that it is superior.

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @08:30AM (#9986798)
My theory is very simple: more math less women.

This predicts the % of women in academic physics relative to academic bio, IT relative to HR, and quite a bit more. Anyone who disagrees is welcome to try and find cases where this theory fails to fit the data.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @08:40AM (#9986826)
Except it doesn't, unless you squint really, really hard. Finding cases where this theory fails to fit the data is too easy: women participation in physics and math is growing and is far higher than in CS, where it's declining; more women go into biology than into history (women participation in history and physics is quite similar). It also doesn't explain the low women participation in politics (again, similar to or lower than in physics[1][2]).

That you continue to believe this theory despite so much contradictory evidence is perplexing.

Also, it's a bad theory in that it's a description rather than an explanation. Even supposing it fit the data better than it does, why would it be the case that more math = less women? I mean, no one suggests that men are 5x or 8x better at math than women as direct studies clearly show it is not the case. So even if it were a good description (though it isn't) it is not the least bit explanatory. The ultimate explanation of such a theory is still likely to be social.

It is probably true that women, on average, are more empathic than men, and therefore would prefer professions that involve working with others more, but that difference is nowhere near the actual participation rates that we find, so it can't be a major explanation.

A "theory" which fits the data -- on average, of course -- much, much better than yours would be "women choose professions with less money and less prestige". This "theory" is surprisingly predictive because, amazingly enough, women choose to be paid less money than men even while doing the very same job, so it must be right.

[1]: http://www.unwomen.org/en/what-we-do/leadership-and-politica...

[2]: https://www.american.edu/spa/wpi/upload/2012-Men-Rule-Report...

Re:Django Girls one year later by sfk on Saturday August 01, 2015 @07:23AM (#9986656)
Here it is:

http://philip.greenspun.com/careers/women-in-science

Obviously this is about averages, there will be several women here to whom the theory does not apply.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @08:09AM (#9986751)
That's your theory? Women participation in math and the physical sciences is extremely high relative to their participation in software, where it used to be much, much higher (same as in other professions) and is constantly declining. That decline actually coincides with increasing, rather than decreasing salaries and prestige, and is not due to increasing work hours (that are lower in software than in medicine and law, where women participation is high). Software is a very clear anomaly.

That theory is even wrong for science in general, especially considering that women participation has been increasing, and that some scientific fields -- that are just as demanding -- are actually dominated by women.

Re:Django Girls one year later by pron on Saturday August 01, 2015 @06:19AM (#9986537)
> There is plenty of reason to study them; they provide validation/rejection of our theories.

That's not always enough to fund sufficient research, especially as those avenues of research show no more promise in uncovering a more fundamental model or validating/rejecting theories better than the ones that are currently explored.

> yet fields with no power difference have disparities which directly mirror the disparities you claim are driven by "power", that suggests there is an underlying theory that doesn't use power at all and which predicts more things.

If only you were right about the facts then your (unique, I must say) theory -- which I can only guess at[1] because you won't state it clearly -- would merit a closer look, but you're not. You are equating existence of bias with existence of bias, but with no regard to effect size. Here are the facts: http://www.randalolson.com/2014/06/14/percentage-of-bachelor...

CS participation was the same or higher than math and physics, and now it's significantly lower. The difference today between participation in math and physics vs CS (and engineering) is far higher than the relative difference between biology and physics. Indeed, it is higher than the relative difference between English and physics.

The social power theory is currently the only theory that has enough evidence to support it, and it gathers more evidence every day, which is why it is the (current, at least) consensus of the scientific community. Your theory (whatever it is), describes neither the observed differences in academia nor in the industry and runs counter to the body of knowledge we've accumulated about power dynamics, group behavior etc.

[1]: If your theory claims innate ability is a major cause of the difference, then it has been disproven. No differences in innate ability were found to be even in the ballpark of the participation difference we see in society. At this point in time, we can say with a high degree of confidence that innate ability is not a major explanation. Even evolutionary theories would reject ability-based explanations, as there is little reason for such huge differences in ability (way beyond difference in physical strength) in a species where the two sexes play almost identical roles (compared to, say, insects). If you have another theory, please state it.

Re:Django Girls one year later by yummyfajitas on Saturday August 01, 2015 @08:50AM (#9986850)
CS participation was the same or higher than math and physics, and now it's significantly lower.

Yes - and this contradicts belorn's theory.

My theory is very simple: more math less women, with other factors being secondary. Here is Scott Alexander doing some pretty good regressions (r=-0.82, p=0.0003): http://slatestarcodex.com/2015/01/24/perceptions-of-required...

There is lots of evidence of innate ability playing a significant role in this. Here are some papers:

http://www.ams.org/notices/201201/rtx120100010p.pdf http://itp.wceruw.org/Hyde%20Science%2008.pdf http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3057475/pdf/nihm...

(Note that putting agreement with me into an abstract is career suicide, so you'll actually need to skip ahead to the data tables.)

My theory predicts % of women in academia in bio vs physics, % women in IT office work vs softer office work, % women in quant trading vs back office, and many more. Do you have any evidence at all which contradicts this?

Re:Django Girls one year later by pron on Saturday August 01, 2015 @09:32AM (#9986962)
> Yes - and this contradicts belorn's theory.

Scientific consensus also disagrees with that theory. We believe that the reason is related to power disparity not simply being a minority group.

> Here is Scott Alexander doing some pretty good regressions (r=-0.82, p=0.0003).

Except that it explains little, as women participation in all fields is rising. So while it may always be the case that you'll find that correlation with mathematical ability, it matters little, as it explains smaller and smaller differences. In fact, a low analytical writing score may correlate better than a high math score. Where things get weird is in CS, and worse: the software industry.

> There is lots of evidence of innate ability playing a significant role in this. Here are some papers...

That's not what the papers show at all. The greatest difference I spotted was twice as many boys as girls over the 99th percentile. We are talking 5x-10x participation gap in software.