Implicit bias

There have been a number of blog posts in the last week or so about a study that looked at implicit (rather than explicit) discrimination in hiring practices. Both Jenny Rohm and Athene Donald have had interesting things to say.

The abstract says

Despite efforts to recruit and retain more women, a stark gender disparity persists within academic science. Abundant research has demonstrated gender bias in many demographic groups, but has
yet to experimentally investigate whether science faculty exhibit a bias against female students that could contribute to the gender disparity in academic science. In a randomized double-blind study (n = 127), science faculty from research-intensive universities rated the application materials of a student—who was randomly assigned either a male or female name—for a laboratory manager position. Faculty participants rated the male applicant as significantly more competent and hireable than the (identical) female applicant. These participants also selected a higher starting salary and offered more career mentoring to the male applicant. The gender of the faculty participants did not affect responses, such that female and male faculty were equally likely to exhibit bias against the female student. Mediation analyses indicated that the female student was less likely to be hired because she was viewed as less competent.

This is scary stuff. The researchers explicitly make the point that they don’t think it’s due to explicit bias:

… we are not suggesting that these biases are intentional or stem from a conscious desire to impede the progress of women in science. Past studies indicate that people’s behavior is shaped by implicit or unintended biases, stemming from repeated exposure to pervasive cultural stereotypes that portray women as less competent but simultaneously emphasize their warmth and likeability compared with men.

So even though I have no explicit bias against women, it’s highly likely that I have an implicit bias. As I said, scary.

What can we do about this? There are some fairly obvious things, but I simply don’t know whether they are enough.

First, be aware of the possibility of bias, and look out for it. Following a link trail from Athene Donald’s blog led me to a really useful definition of discrimination used by Shara Yurkiewicz:

A preference is discrimination when:

1) decisions such as those about hiring people and setting their pay rate are based on generalizations about the demographic groups to which individuals belong

2) individuals have no control over the group to which they belong – and it is apparent from their appearance.

3) it is nearly impossible to predict how an individual will do the job based on the group to which he or she belongs.

In fact I had an argument along these lines with the speaker at a dinner I went to last night. He used the old joke about men’s and women’s thought processes being completely different. I found the joke marginally offensive, and told him so afterwards. He claimed that it was based on facts — men, on the whole, are more analytic, less touchy feely, and so on. I said on average, maybe, but what we have is two overlapping distributions. He agreed, but the root of the problem is that we disagree over the extent of the overlap. I am firmly of the opinion that the distributions are both pretty wide, with a large overlap, which according to the definition above puts us firmly in discrimination territory.


risk management

Are VaRs normal?

An article in the FT’s recent special report on Finance and Insurance started from the premise that VaR models were a significant factor in landing banks with huge losses in the wake of the collapse of the US housing market, and went on to discuss how new models are being developed to overcome some of their limitations. Part of the point is valid — there were many models that didn’t predict the collapse. But the article is positively misleading in places. For instance, it implies that VaR models are based on the normal distribution:

VaR models forecast profit and loss, at a certain confidence level, based on a bell-shaped, or “normal”, distribution of probabilities.

In fact Value at Risk, or VaR, is a statistical estimate that can be based on any distribution. And it’s pretty obvious that for many financial applications a normal distribution would be inappropriate. The people who develop these risk models are pretty bright, and that won’t have escaped them. The real problem is that it’s difficult to work out what would be a good distribution to use — or, more accurately, it’s difficult to parameterise the distribution. To get an accurate figure for a VaR you need to know the shape of the distribution out in the tails. And for that, you need data. But by definition, the situations out in the tails aren’t encountered very often, so there’s not much data. And that applies whatever the distribution you’re using. So simply moving away from the normal distribution to something a bit more sexy isn’t necessarily going to make a huge difference to the accuracy of the models.

The article goes on to discuss the use of Monte Carlo models to calculate VaR. Monte Carlo models are useful if the mathematics of the distribution you are using don’t lend themselves to simple analytic solutions. But they don’t stop you having to know the shape of the distribution out in the tails. So they do help extend the range of distributions that can usefully be used, but it’s still a VaR model.

And that’s another problem entirely. VaR, like any other statistical estimate (such as mean, median or variance) is just a single number that summarises one aspect of a complex situation. Given a probability and a time period, it gives you a threshold value for your loss (or profit —  but in risk management applications, it’s usually the loss that’s of interest). So you can say, for instance, that there’s a .5% chance that, over a period of one year, your loss will be £100m or more. But it doesn’t tell you how much more than the threshold value your loss could be — £200m? £2bn?

And it’s a statistical estimate, too. .5% may seem very unlikely, but it can happen.

I wouldn’t disagree that a reliance on VaR models contributed to banks’ losses, but I would express it more as an over-reliance on models, full stop. It’s really difficult to model things out in the tails, whatever type of model you are using.


Models and modellers

On 1 March I gave the Worshipful Company of Actuaries lecture at Heriot-Watt University. Here’s the abstract:

Being an actuary nowadays is all about modelling, and in this lecture I’ll discuss how we should go about it. We all know that all models are wrong but some are useful – what does this mean in practice? And what have sheep and elephants got to do with it? Along the way I’ll also consider some of the ways in which the actuarial profession is changing now and is likely to change in the future, and what you should do about it.

And here’s what I said.


Are boom periods bad for us?

Apparently, in the USA at least, death rates rise during periods of economic expansion and fall during economic downturns. I don’t know whether this holds in the UK as well. One possible reason for this is that when people feel well off they eat and drink more (and more unhealthily). Another is that people drive more, so there are more car accidents.

Yet another, according to a recent study, is that in good times nursing homes find it more difficult to hire care assistants because of the low wages.

Modelling future mortality rates is seriously difficult.


Does size matter?

Over the last few months there have been several interesting pieces about innovation, the size of companies, and other rather loosely connected topics.

Back in December, the Schumpeter column in the Economist reviewed an article arguing that large firms are often more innovative than small ones. This seems counter-intuitive – surely small companies are nimble and creative, and large ones are stultified by bureaucracy. But it turns out that there are good reasons why intuition may be wrong here, although Schumpeter thought that there are limits to the argument: large firms may be innovative, but it’s usually incremental innovation rather than disruptive innovation.

This all fits in quite well with an observation a friend and I made recently when discussing discrimination in employment. You might think that there will be less diversity in large, established companies than in small companies, especially high-tech startups. But in our experience the converse is true. We thought it was probably because large companies have better processes in place, are more conscious of the issues and are less likely to hire on the basis of existing friendships. Also, of course, they have larger workforces, so can reflect population diversity more easily than a small company in which every person constitutes say, 10% of the employees.

Conventional wisdom also has it that most job creation happens in small companies. Well, up to a point. Recent research has found that in fact once you control for the age of the firm there is no systematic relationship between firm size and employment growth, in the USA at least. Small firms tend to be younger — if they are older, then they are less successful (else they’d be bigger). (I can’t resist: repeat after me, correlation is not causation).

The Economist has a leader this week that discusses the contrast between large and small firms. It argues that, on the whole, economies relying more on small firms have been less successful than those with more large firms (think northern and southern Europe). Apparently productivity is much greater in large firms, at least partly because of economies of scale. Don’t have special regulatory and fiscal breaks for small firms, the leader argues, but go for growth instead.



Blogging hiatus

There’s been a hiatus in my blog recently. It was due to life getting very very busy and a bit out of control, but I think it’s back to normal now. It was all out of control in a very good way, incidentally!

So something approaching normal service will now resume. We apologise for any inconvenience caused.

risk management

Unintended consequences

Facebook bans at work are apparently linked to increased security breaches. It seems that strict policies on social networking sites are “actually forcing users to access non-trusted sites and use tech devices that are not monitored or controlled by the company security program.” People are very adaptable, and often very determined. If you stop them doing something one way, they’ll find another. Computer security is really difficult, as it’s by no means a matter only of technology: human nature is a major factor, and often more easily predicted with the benefit of hindsight.

For instance, Bruce Schneier points out that if something’s protected with heavy security, it’s obviously worth stealing. It’s the converse of Poe’s The Purloined Letter, in which the best hiding place is in full view. Does this apply to computer systems?


Are we nearly there yet?

The title of this blog is a shameless crib from a recent blog of Athene Donald’s, in which she discusses the Equality Challenge Unit‘s annual survey of statistical information about staff and students in UK universities.

[…] overall 76% of professors are white and male. Such a lack of diversity cannot be healthy. The numbers of BME (black and minority ethnic) staff across the board, male or female, is truly dismal. A mere 5.3% of academic staff are non-white UK nationals and there are a further 6.6% of non-UK BME staff members.

More girls than boys go to university, although this gap is slowly decreasing (from 14.6 to 13.2% over the period from 2003/4). In some subjects the disparity is huge:  80.6% are girls in subjects allied to medicine, 76.6% in veterinary sciences, and even in biological sciences the percentage is 62.9%.

She concludes with the interesting question:

So, we should be asking ourselves, not only ‘are we nearly there?’, but where is the ‘there’ we are trying to reach. Is the ideal a 50:50 split between the genders at all levels and for all subjects, or do we believe that this is a) impossible or b) undesirable – or even c) irrelevant as a metric.

Meanwhile, it’s fairly obvious from other sources that we’re not nearly there, for any reasonable definition of “there”, even leaving aside the obvious matters of the gender pay gap and the dearth of women in top jobs.

Personally, I’m not a huge sports fan. Well, not really a sports fan at all, to be honest. I do participate at grass roots level (let’s hear it for parkrun), but I don’t really follow or even watch sport. But I’d like to be able to not watch women’s sport on an equal footing to men’s. From Zoe Williams:

A young female rower told me two years ago that the big scandal of the way women were treated in UK sport was best illustrated by netball: it was never covered by the media, even though we were among the best in the world.

As host nations of the Olympics, we could have nominated it as one of our four new events. Instead, we chose women’s boxing: no spectator base, no foothold in schools, no realistic chance of it catching on, but you wanted equality, ladies? Here, take a punch in the face.

There are huge numbers of sports fans, but they don’t see many women. But then, people listening to the Today Show don’t hear many women, those watching Question Time don’t see many, and people reading newspapers don’t read women’s words, according to recent research. Women are seriously under-represented in the media.

And it gets worse. As I wrote last month, there’s a lot of misogyny around, and a number of women wrote about what they encountered. A week or so ago, Nick Cohen wrote a piece on the subject, and as described by Ellie Mae O’Hagan

Almost as soon as the piece was published, “Nick Cohen” started trending on Twitter. Clicking on the topic revealed scores of men and women sharing and praising his article; congratulating him for “nailing” the subject.

Why, she asks, did Nick Cohen trend on Twitter?

After all, it didn’t trend on Twitter when women pointed it out; and if I remember rightly, a great deal of respondents told us to stop being so weak. […] How strange, then, that Cohen’s piece should be the subject of such adulation. How unfathomable it is that his opinion should be lauded more than those for whom misogyny is a lived experience. It seems, as one Twitter user put it to me, that when “feminist women call sexism they are portrayed as killjoys; when feminist men do it, they are portrayed as white knights riding to the aid of defenceless women.”

There’s some progress, though. Well, maybe. Hamley’s has stopped colour coding its floors pink and blue for girls and boys. That’s bound to make a difference. Isn’t it?



Interesting links

I found these interesting:

  1. Kaprekar’s constant — not everything has to be useful to be appealing and fun.
  2. Apparently the Roman Empire was more equal than the USA, while in Britain income inequality rose faster between 1975 and 2008 than in any other OECD member country.
  3. How to get your keys back if you drop them down a drain.
  4. Talking about big numbers
  5. The UK opens up NHS data, and the EU announces an ‘open by default‘ position for public sector information.

Correlation is not causation, part 999

A few weeks ago the Economist’s blog had a piece with the tag line “How increases in computing power have driven higher share turnover”. It shows a nice chart with two lines rising inexorably upwards, pretty close together, one representing the transistor count in integrated circuits from 1947 to date, and the other shares traded on the New York Stock Exchange over the same period.

Computing power has increased some 600-fold over the past 15 years […] This advancement has facilitated the ability to trade ever-larger volumes of shares.

Whenever I read something like this, my knee-jerk reaction is “correlation is not causation”. Just because two phenomena behave in roughly the same way, it doesn’t mean that one of them is causing the other. One of the better known examples of this is the strong statistical association between the annual changes in the S&P 500 stock index and butter production in Bangladesh. Admittedly it’s plausible that increased computing power has contributed to higher share turnover, but “driven” seems rather strong.

I stick by my knee-jerk reaction, but after discussing it with a friend I think there’s something even less satisfactory going on. The chart showing these two inexorably rising lines uses a logarithmic scale. And the lines are actually pretty divergent from about 1970 onwards. What this means is that the rate of growth isn’t even the same. This is a very tenuous hook indeed on which to hang a conclusion of causality.