I think that UN manipulating it's own index is not culture wars even if the index is related to gender. Let me know if I am wrong.
Human development
The Gender Development Index (GDI), along with its more famous sibling Human Development Index (HDI) is a an index published annually by UN's agency, the United Nations Development Programme (UNDP). Whether an index is manipulated or not can be judged only against a precise definition of what the index claims to be measuring. So how do you measure human development? Whatever you do, you will never capture all nuances of the real world - you will have to simplify. The UNDP puts it this way:
The Human Development Index (HDI) was created to emphasize that people and their capabilities should be the ultimate criteria for assessing the development of a country, not economic growth alone.
So the UNDP defines the Human Development Index as a geometric mean of three dimensions represented by four indices:
| Dimension | Index |
|---|---|
| Long and healthy life | Life expectancy at birth (years) |
| Knowledge | Expected years of schooling (years) |
| Mean years of schooling (years) | |
| Decent standard of living | Gross National Income (GNI) per capita (2017 PPP$) |
Source: https://hdr.undp.org/data-center/human-development-index#/indicies/HDI
Gender Development
So far so good. Next, on it's website the Gender Development Index (GDI) is defined like this:
GDI measures gender inequalities in achievement in three basic dimensions of human development: health, measured by female and male life expectancy at birth; education, measured by female and male expected years of schooling for children and female and male mean years of schooling for adults ages 25 years and older; and command over economic resources, measured by female and male estimated earned income.
Source: https://hdr.undp.org/gender-development-index#/indicies/GDI
While in the actual report HDI it is simply defined as a ratio of female to male HDI values:
Definitions - Gender Development Index: Ratio of female to male HDI values.
Source: https://hdr.undp.org/system/files/documents/global-report-document/hdr2021-22pdf_1.pdf
Let's look, for instance, at the Gender Development Index of United Kingdom. The value 0.987 means that despite longer life and more education, in UK, females are less developed than males.
| Dimension | Index | Female value | Male value |
|---|---|---|---|
| Long and healthy life | Life expectancy at birth (years) | 82.2 | 78.7 |
| Knowledge | Expected years of schooling (years) | 17.8 | 16.8 |
| Mean years of schooling (years) | 13.4 | 13.4 | |
| Decent standard of living | Gross National Income (GNI) per capita (2017 PPP$) | 37,374 | 53,265 |
Source: https://hdr.undp.org/system/files/documents/global-report-document/hdr2021-22pdf_1.pdf
Wait, what?? What does it mean that females in UK have command over economic resources of post Soviet Estonia (GNI Estonia=38,048) while males in UK have command over economic resources of EU leader Germany (GNI Germany=54,534)?
The manipulation
The UNDP calculates separate command over economic resources for females and males, as a product of the actual Gross National Income (GNI) and two indices: female and male shares of the economically active population (the non-adjusted employment gap) and the ratio of the female to male wage in all sectors (the non-adjusted wage gap).
The UNDP provides this simple example about Mauritania:
Gross National Income per capita of Mauritania (2017 PPP $) = 5,075
| Indicator | Female value | Male value |
|---|---|---|
| Wage ratio (female/male) | 0.8 | 0.8 |
| Share of economically active population | 0.307 | 0.693 |
| Share of population | 0.51016 | 0.48984 |
| Gross national income per capita (2017 PPP $) | 2,604 | 7,650 |
According to this index, males in Mauritania enjoy the command over economic resources of Viet Nam (GNI Viet Nam=7,867) while females in Mauritania suffer the command over economic resources of Haiti (GNI Haiti=2,847).
Let's be honest here: this is total bullshit. There are two reasons why you cannot use raw employment gap and raw wage gap for calculating the command over economic resources:
Argument 1
Bread winners share income with their families. This is a no brainer. All over the world, men are expected to fulfil their gender role as a bread winer. This does not mean that they keep the pay check for themselves while their wives and children starve to death. Imagine this scenario: a poor father from India travels to Qatar where he labours in deadly conditions, so that his family can live a slightly better life. According to UNDP, he just became more developed, while the standard of living his wife is exactly zero.
Argument 2
Governments redistribute wealth. This is a no brainer too. One's command over economic resources and standard of living is not equal to ones pay check. There are social programs, pensions, public infrastructure. Even if you have never earned a pay check yourself, you can take a public transport on a public road to the next public hospital. Judging by the Tax Freedom Day, states around the world redistribute 30% to 50% of all income. And while men pay most of the taxis (obviously, they have higher wages) women receive most of the subsidies (obviously, they have lover wages). But according the UNDP, women in India (female GNI 2,277) suffer in schools and hospitals of the war-torn Rwanda, while men in India (male GNI 10,633) enjoy the infrastructure and social security of the 5-times more prosperous Turkey.
Don't get me wrong, the employment gap and pay gap are not irrelevant for the standard of living and command over economic resources. Pensions and social security schemes mostly do not respect the shared family income and as a result the partner doing less paid work - usually a women - gets lower pension, unemployment benefit etc. What's worse, the non-working partner is severely disadvantaged in case of divorce or break up. But while this has an impact on each gender's standard of living it certainly does not define 100% of that value.
Argument 3
You may argue that the command over economic resources measured by estimated earned income is some kind of proxy for all other disadvantages women face in society. But do you remember what I said in the beginning?
Whether an index is manipulated or not can be judged only against a precise definition of what the index claims to be measuring.
The HDI measures "people and their capabilities" and the GDI is a ratio of these capabilities measured separately for men and women. The economic dimension of the GDI is supposed to be standard of living or command over economic resources - neither of which can be represented by earned income alone.
The taboo
Wikipedia says: "For most countries, the earned-income gap accounts for more than 90% of the gender penalty." (I have not verified this.) This is important, because when we look at the other two dimensions it becomes clear that while men have shorter and less health lives they also increasingly fall behind in mean and expected years of schooling. Without the misrepresentation of the command over economic resources value, the index would show something very uncomfortable: that according to UN's own definition of Human Development men are the less developed gender.
PS: Is there a way to give those tables some borders and padding?
Jump in the discussion.
No email address required.
Notes -
Policy affects murder rates and accidents. Change policing/education/punishment/lead exposure/choose your own adventure, and you get different crime outcomes. Choosing certain sets of policies that disproportionately disadvantage men damages gender equity and should be included in any metric attempting to represent it.
You're assuming your conclusion: Iceland has achieved perfect gender equity, therefore it has achieved perfect gender equity.
What error, exactly, do you think is being minimized? You propose that we have two unknown distributions (biological contributions to lifespan gap and and gender inequity contributions, by country); we only get observations of their sum. You've provided no justification for assuming that the mean of one distribution (biological factors) is the mean of their sum. In fact, the only way that's mathematically possible is if the mean of the gender inequity distribution is zero (i.e. you know a priori that all the instances of anti-female discrimination are perfectly balanced by instances of anti-male discrimination). That's certainly a position you can take, but I can say with a high degree of certainty that that's not the position that the people at the UN hold.
But let's roll with your theory: you somehow know the social inequity distribution has a mean of 0 years, and the biological distribution has a mean equal to 5 years. What, then, can we say about the social inequity variable for one country given the observed variable? Still very little; we still don't know what the unknown distributions are, or even if they're normal, let alone what their standard deviations are. You can't even say that large sum deviations from 5 years hint at large social inequity values, because it could be driven entirely by biological deviations.
The only world where this model offers a good measurement of the gender inequity variable is one where the mean of the biological distribution is 5 years with small deviations compared to the social inequity distribution, which would have a mean of zero (no net discrimination) and account for the large majority of the variation in the sum distribution. Iceland does, it turns out, have to be a relative hellhole for women compared to Pakistan and Sudan.
I don't think that's the world that exists, and I strongly suspect it's not the world the folks making the index at the UN think exists either. Do you?
Iceland was your example, was it not? And, no, I am not. Were I making that argument, I would have said that the estimate should be 3. The point is that we don’t know, so the issue is, given what we know today, what is our best estimate of the average biological component across countries, races, etc? Our best estimate isn't zero, nor particularly close to zero.
The usual: the extent to which changes in the metric reflect actual changes in what the metric is attempting to assess. Both false highs and false lows, but of course there is usually a tradeoff between them. Were the data binary, using something like the area under the ROC curve. The data here is continuous, and supposedly there is an analogous method for use with continuous variables, but I don't know enough about the topic to say if area under the ROC curve is appropriate here. I know that many other methods exists, but that is all I know about them.
Of course you can say it hints at large social inequity values, and of course it could be driven entirely by biological deviations. Those are not mutually exclusive statements.
As for the rest, I really don't understand your point. Perhaps I misunderstand you, but wouldn't your logic be the same using 4 years, rather than five? And 3 years? And 3 months? Indeed, everything but zero, which we know is incorrect? If your argument leads to the conclusion that the only legitimate estimate is one that we know is wrong, it seems to me that something has gone awry. I also think you might be using a number that is meant to be an average across countries, and assuming that it claims that that is the average in every country.
Also, suppose you saw a country in which women lived 10 years less than men. Would you not stroke your chin and say, "hm, I suspect that something is amiss in that country, because women usually live longer than men"? Isn’t 5 years a rule of thumb for when you should start scratching?
Finally, I think we are losing sight of the point of the GDI, which is to ensure that progress on the HDI does not cause people to overlook the fact that sometimes such progress is not shared as broadly as it could be.
This criticism also applies to 0 years. It's simply impossible to separate the biological component from the gender inequity component with only observations of the sum. It's not impossible if you bring in other data, but the GDI doesn't cite any data sources in its methodology to justify the 5 years. You can assign whatever probability you want to the UN Development Program having a crack team behind the scenes doing secret HBD research.
There's a certain vacillation here. On one hand, the GFI is a highly technical, synthetic metric only to be used by experts in conjunction with the HDI in measuring temporal trends and doing international comparisons. If that's the case, though, why apply the unevidenced fudge factor? If it makes some countries look like they favor women, why try to correct for it at all? Experts would know that it's driven by the difference in male and female life expectancy.
So suppose I am a teacher, and I am biased against men. So here is how I grade essays: First, I grade them blindly. Then, after removing the blinds on the names, I enter the grades in a spreadsheet. Next, I reduce the scores of male students by an average of five points, with a distribution identical to the distribution of differences in earnings by gender. I enter the new scores in my grade book. After the scandal is revealed, the school adds five points to every male's score. Are you saying that the result would not likely be a more accurate representation of the actual scores than was my original list?
??? How do you know this? Are you privy to the no doubt voluminous documentation that has doubtless been generated over the years? Have you perused the numerous papers available on Google Scholar which discuss the GDI to see if any assess the methodology in question?
Or, you could do what they do, and those same experts, and outsiders who use the GDI, are free to delete the adjustment if they want to. As I have said several times, if including the adjustment results in an index which more accurately reflects actual conditions than an index using unadjusted numbers, then that is a good reason to use it.
Here you are assuming you know something concrete about the teacher bias. When the scandal comes to light, that's the mean teacher bias factor being made known to the school. So, yes, the adjustment the school makes improves the accuracy of the official grades, but the point is that you're assuming they know the mean bias.
The equivalent here would be the school observing that girls scored higher than boys, assuming that it was entirely due to teacher bias, and then adjusting the boys' distribution to have the same mean as the girls'.
I've read through the methodology and technical notes that's provided for the GDI. Perhaps they have some unpublicized documentation providing more justification for taking the average lifespan difference as the biological one (in other words, that there is no net gender-based inequity in health); you're free to look for it if you want. Usually, though, if you're going to add an arbitrary fudge factor in a paper, you describe its motivations and provide evidence for it when you introduce it.
And, in a thread analyzing the construction of the GDI, I'm also free to criticize how it's constructed.
But, were you not arguing that it was impossible, even we knew the mean?
Except that neither the hypothetical nor the GDI assumes that the difference is entirely due to the factor being controlled for. The whole point of the GDI is to try to figure out what the difference would be, were the biological effect zero. And in neither the hypothetical nor the GDI is the outcome that the male and female distribution have the same mean.
You keep using this term, "arbitrary," despite it clearly not being arbitrary. It is based on observations of relevant data. It might nevertheless be , or too high, or too low, or based on assumptions that are subject to dispute. But they are not arbitrary.
But, we have already discussed that. The only thing I don’t understand is why you think that it is mathematically impossible to make the adjustment.
The only reference I made to impossible was:
That's coming from a position of total ignorance about the biological component of the observations and only having access to the observed values.
Knowing something about the biological component does give you information about the gender inequity component, given their sums. If you know the biological component mean, for instance, you also know the gender inequity component mean. I'm not sure what else you get (maybe an expert in statistics can weigh in if they're reading this deep into the thread). Here, though, we don't know a priori the biological mean contribution: all we have is a vague sense that there is a biological component that favors women. It's a leap to go from that to "the average difference in life expectancy is entirely due to biological differences." If the actual biological component were 3 years as opposed to 5 years, we end up understating female privilege; the same could very well hold in the opposite direction if the actual biological difference were 7 years, as in Japan. There's just no basis to say what the actual biological difference is either way, given only the sum of the component differences.
It does, though.
On the contrary, by subtracting the observed mean difference in lifespan, by construction the GDI results in the same mean gender-specific component values. If you take the UN's data and calculate their adjusted individual health components for men and women for each country, the average for women is 0.79. And the average for men turns out to be... 0.79. This means the average "health GDI" (the ratio of the two) is 1. That is to say, men and women have apparently achieved health equity on average worldwide, if you assume that the observed average difference in lifespan is exactly the same as the average biological component to lifespan.
Calculations: https://pastebin.com/raw/Hkiejg4h
Source spread sheet: https://hdr.undp.org/sites/default/files/2021-22_HDR/HDR21-22_Statistical_Annex_GDI_Table.xlsx
(Apologies for the pastebin CSV; I would share the sheet directly but want to minimize linkage to my real life identity. You can go into Google Sheets and "Paste Special > CSV as columns" to recreate it.)
Oh, I meant to refer to your initial statement that "the only way that's mathematically possible"
Again, this seems to be an argument the difficulty in achieving perfect accuracy, which we have talked about before. We know the biological component is not zero, but to pretend that it is zero simply because it is impossible to know precisely what it is does not seem conducive to assessing policy outcomes.
Sorry, I was distracted by my hypothetical, and by being a bit imprecise. In my hypothetical, there is no reason for the male and female means to be the same, because the average of a five point deduction was not derived from the underlying data. And of course you are right, because if you deduct the average difference from the higher gender, the resulting means will be the same?
But, who cares? When I said, "The whole point of the GDI is to try to figure out what the difference would be, were the biological effect zero," I meant the difference in each country. And when I said, re my hypothetical, "Are you saying that the result would not likely be a more accurate representation of the actual scores than was my original list?", I meant each score.
I think I now understand that your statistical arguments have been about global numbers, while I have been talking all along about scores within individual countries, because that is how the GDI is used: individual GDI scores are compared with individual HDI scores.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
If you're asking "could the index still be useless for forming policy if it used 4 years or 3 months?" Sure. If you have no way to know that you got it right no matter what value you use, then you have no way to know you got it right, period.
I'd say that 10 years less is a clear statistical outlier. Just "less than 5" doesn't make something a statistical outlier.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link