Comment by facepalm | Hacker News Reader

Comment by facepalm | Hacker News Reader

Parent Full thread

Well I don't have that much time, so for now I only looked at pictures with people in them they have posted since December 1.

Simply counting people yields 29% white women, 19% women of color, 21% men of color, 30% white men.

Not a very exact science, though - I left out groups above a certain size (for example picture from Anti-Trump demonstration or MLKday), and in some cases I couldn't recognize the people. Many white men come from office shots where they linger in the background, whereas there are many tweets explicitly featuring female or black engineers. It seems by only looking at their "media timeline" I also missed photos like this one: https://twitter.com/Every28HoursPla/status/83166688771703193... (which they retweeted).

I'll try to find time for a better "analysis", ideally including texts.

Compared to last time there seem to be now more posts boasting technology at Google. For example there were several about Tensor Flow, all featuring the same white guy (I counted him for every instance).

I couldn't find the time when I last posted about lifeatgoogle, would have liked to look at their tweets from around then.

For comparison employee stats from 2014: http://mashable.com/2014/05/28/google-employee-demographics/... - 70% male, 91% white or asian.

So what's your conclusion?

This time around there were more white men, I think - in part because of some specials like Tensor Flow, or a picture from an actual office.

It is still biased against white men (if it is supposed to reflect the actual distribution of Google employees), but not as extreme as last time. I really would like to find the date of my last comment about it. Also perhaps simply more data is needed - a single picture with several people could shift the results here, because I checked only pics from 2.5 months.

Also better methodology needed, this was just a quick shot looking into one simple metric.

You're again, falling victim to confirmation bias by rejecting the best quantification you've provided yet.

Lets examine the data you've provided - 29% white women, 19% women of color, 21% men of color, 30% white men.

This data is entirely in line with the demographics of the US. About 50/50 on gender and 60% white. In fact, given google's global hiring reach, these figures are actually biased towards white people - while about spot on for gender. This entirely contradicts the point you were originally pointing to this twitter feed as confirmation of.

If anything, this thread has just made me newly impressed with Google's approach to inclusion. It's also really, really obvious that the person you are responding to isn't able to reconcile seeing PoC and women with their own world view.

Huh wtf - what does seeing women and PoCs have to do with a "world view"? You think I am not aware that women and PoCs exist?

Based on your quantification, there's now more evidence for the claim that you perceive an unbiased sample of people to be biased towards minorities and women, than there is evidence against that claim. It seems like you're struggling to reconcile this bias with the quantification of it.

HN won't give me a reply button further down, and I want to go to bed, so I am replying here:

Yes, it reflects demographics of the US, but not demographics of tech or demographics of Google employees. So the account definitely doesn't reflect life at Google in an unbiased way.

Also, I tried to err on the side of counting too many whites. For example I counted this screenshot from an animation movie as two whites: https://twitter.com/lifeatgoogle/status/824649101069455361 I counted the blurry people in the background of this office: https://twitter.com/lifeatgoogle/status/817466019526610947 but I only counted 2 PoCs here despite the further pictures with more https://twitter.com/lifeatgoogle/status/822528000646381568 I also missed a lot of pictures because I didn't realize the retweets wouldn't be in the "media list".

I just made it up on the fly for a quick, simple metric. It would be better to decide beforehand what counts, for example if the person should be the item of a news story, should be presented as an engineer, stuff like that. And a longer time. I think I had 150 people, so the animation picture alone accounted for more than 1% of the final count of white people. As I said, some office shots greatly raised the white people count, counting only people who were subject of major stories would have lowered the percentage a lot.

Maybe you are jumping to conclusions because they confirm your beliefs?

Actually I had no idea what the demographic percentages of the US were before I looked them up.

Your original point was that you thought Google wasn't interested in hiring male engineers anymore. The current demographics of Google are irrelevant to Google's hiring strategy. Why would they be?

You're not only displaying confirmation bias in the way you are trying to undermine the clearest quantification you have access to, but you're also avoiding your original statement.

You initially presented the lifeatgoogle twitter feed as evidence, and you had no quantifiable evidence that it was biased. Now that you do have a quantification, you're walking back the importance of that evidence. Perhaps you're doing this so you can maintain your poorly quantified view?

If only you gave as much latitude to other people.

My claim about them not presenting any white males was from another time - do you recall by any chance when it was? As I explained, the sample size I used now was small, and a few pics can have made a big difference. I think when I made the claim there were different pictures, that is why I said more data should be looked at (to reduce random variation). I probably wouldn't have made the claim about the current timeline.

And again - I only used one simple metric, which already shows bias (it doesn't represent the actual demographics of Google employees). You assume now that metric is conclusive because it fits your conviction. By looking into more aspects the picture would be more clear.

And where do I not give latitude to other people?

>You assume now that metric is conclusive because it fits your conviction.

No, you're now walking back your claims because the one metric you have quantified doesn't fit your conviction. At the time you claimed that by simply looking at the feed you could tell it was biased. You initially made the claim that the lifeatgoogle twitter was related evidence to your claim that google wasn't interested in hiring men anymore. I'm not silly enough to claim that a brief perusal of a twitter account can be extrapolated into a claim about a company's hiring strategy.

Thanks to your quantification, we have some evidence to suggest the lifeatgoogle twitter is fairly representative of population demographics along gender and racial lines, at least in the US. I don't see the data as conclusive of anything more than that.

I have found our original discussion, it was 57 days ago. So let's do another analysis of their timeline two months ago (December+November, for example). But this time, you tell me how to categorize their tweets.

For example, should this picture count? It is not somebody working at Google, just a stock image of somebody working in the food industry: https://twitter.com/lifeatgoogle/status/813410799725215744

And how should pictures like this be counted: https://twitter.com/lifeatgoogle/status/800818789122129926 or this https://twitter.com/lifeatgoogle/status/812002215267139584

Maybe from the last two you can also see why perhaps simply counting people is not the best metric - a single group photo could distort everything.

You mention you work in statistics, so tell me how you would approach the problem?

Also maybe I didn't express it clearly enough, I didn't look at the whole timeline, only the media timeline: https://twitter.com/lifeatgoogle/media (which is media they posted, pics and videos).

I also didn't claim I can tell bias simply by looking, I said it was my impression they don't want to hire white men anymore. If I was convinced I could tell simply by looking, I wouldn't agree to do an analysis.

As I said - I am sure they still hire white men, but that particular account (and some of their other publications) give the impression they are not interested in that anymore.

And they claim to represent Google's demographic, not the demographics of the general population, so contrary to what you say my numbers do show bias against white men. Not as extreme as I perceived two months ago, but still.

If your company is 70% men and you only report about what women are doing, it is revisionist.

Imagine if all stories about WWII would only be about female soldiers. The impression would be women fought and died for our freedom, all the sacrifices of men would be forgotten. That is revisionism.

>I am sure they still hire white men, but that particular account (and some of their other publications) give the impression they are not interested in that anymore.

Really? What evidence do you have to back up your impression? Why do they give you that impression?

>You mention you work in statistics, so tell me how you would approach the problem?

I'd measure first and then debate my conclusions based on that measurement - and in an area I hadn't measured - I wouldn't spout my impressions or claims - they wouldn't be useful.

Like this thing, which you said one day, not fifty-seven days, ago:

>Huh - that Twitter feed obviously is tilted towards only showing women and PoCs as engineers.

The twitter feed isn't about only women, or only PoCs. You did the measurement yourself.

You seem to still be holding onto your impressions with contrary quantified evidence in front of you. Where's the stuff that supports you?

So how would you measure it?

Another test could be if there are stretches of time when the account gives the impression, that is length of tweet chains without any white men.

I have explained various flaws of my little analysis, but you prefer to ignore them and claim I have provided evidence for your opinion. Yet you work as a statistician :-(

I give up for now, unless you come up with a proper measurement suggestion.

You want me to suggest a measurement method for a claim you've made. Unfortunately for you, for claims you make, you are responsible for questions of methodology surrounding the evidence you use to support it.

You've demonstrated that you'll ignore the best evidence you have in order to hold an impression that you intuit.

>Huh - that Twitter feed obviously is tilted towards only showing women and PoCs as engineers.

Wrong. You've measured this and you are now saying that one test was flawed. The claim it is "obvious" therefore makes no sense.

>It is not just a "guess" - you can look at that feed and check for yourself.

You looked at the feed yourself and were unable to provide evidence it was biased.

Reply on news.ycombinator.com