Back to [Archive] Other discussions
An unfinished paper we have been working on. We are not quite sure how to go on with the economic data, as we don't have an income function.
So I've been working on this paper more. First by porting the code to R, and then by running the analyses on every year there is data so to look at the change over time. I have moved all the files to OSF: https://osf.io/dei73/
It is not yet ready for submission to ODP, but not too far either.
I tweeted about the initial results. These are assuming no environmental gains in g. Pretty unlikely. To model env. effects, one can simply add IQ points to the countries depending on their scores. I've been thinking of three environmental models, 25, 50 and 75% catch-up, by simply moving countries' IQ towards the Danish IQ. So e.g. using 50% gains, if the country of origin has IQ 80 and Denmark 100, it would be modeled as 90. This should drastically reduce the expected changes, but not eliminate them completely.
https://twitter.com/KirkegaardEmil/status/518907865655152640
It is not yet ready for submission to ODP, but not too far either.
I tweeted about the initial results. These are assuming no environmental gains in g. Pretty unlikely. To model env. effects, one can simply add IQ points to the countries depending on their scores. I've been thinking of three environmental models, 25, 50 and 75% catch-up, by simply moving countries' IQ towards the Danish IQ. So e.g. using 50% gains, if the country of origin has IQ 80 and Denmark 100, it would be modeled as 90. This should drastically reduce the expected changes, but not eliminate them completely.
https://twitter.com/KirkegaardEmil/status/518907865655152640
I tweeted about the initial results. These are assuming no environmental gains in g. Pretty unlikely. To model env. effects, one can simply add IQ points to the countries depending on their scores. I've been thinking of three environmental models, 25, 50 and 75% catch-up, by simply moving countries' IQ towards the Danish IQ. So e.g. using 50% gains, if the country of origin has IQ 80 and Denmark 100, it would be modeled as 90. This should drastically reduce the expected changes, but not eliminate them completely.
Are you familiar with La Griffe's post,"COGNITIVE DECLINE: THE IRREDUCIBLE
LEGACY OF OPEN BORDERS"?
http://www.lagriffedulion.f2s.com/imm.htm
No, but I prefer not to read posts when the writing style is like that. I know many people like that, but I rather like it dry and to the point. I don't have all day.
As from my skimming it, it simply shows that unrestricted immigration leads to lower mean IQ and also lower fraction of smarties. Sure. One can easily calculate the smart fraction in my data. One can also calculate the dumb fraction (below e.g. 75). The low fraction is increasing fast.
As from my skimming it, it simply shows that unrestricted immigration leads to lower mean IQ and also lower fraction of smarties. Sure. One can easily calculate the smart fraction in my data. One can also calculate the dumb fraction (below e.g. 75). The low fraction is increasing fast.
I looked a little bit more at it. Can I ask you how many observations do you have for inequality variable and demographic variables ? By observations, I meant the number of years. I think if you have enough observations for each, you can try to do a Granger causality test. It requires lot of steps, but that's doable in Stata and R.
A question concerning the first paragraph at page 3. Are you saying that for children who have at least one parent of danish origin, he will be considered danish, no matter what the origin of the other parent is ? If you're interested in trends over time, this way of labeling people can be a confound if the rate of intermarriage is increasing.
Also, at page 4, when you cite Strenze for saying that g is a better predictor than parental SES on the child's SES outcome (when he is adult), it depends on what you are thinking. Correlation or regression ? By predictor, you must mean regression. Because it's generally under this context that this word is used. In Strenze's analysis, you have correlation, not regression, so you must replace "better predictor" by "better correlate".
I can also add that the information about which of parent SES or IQ is the most important is not the most relevant thing here. What is of more significant finding is whether or not IQ-outcome bivariate correlation remains the same (or diminish) after partialling out SES. Looking at Colom & Flores-Mendoza (2007) for example, I know it's true. See also Thienpont & Verleye (2004).
Also, at page 4, when you cite Strenze for saying that g is a better predictor than parental SES on the child's SES outcome (when he is adult), it depends on what you are thinking. Correlation or regression ? By predictor, you must mean regression. Because it's generally under this context that this word is used. In Strenze's analysis, you have correlation, not regression, so you must replace "better predictor" by "better correlate".
I can also add that the information about which of parent SES or IQ is the most important is not the most relevant thing here. What is of more significant finding is whether or not IQ-outcome bivariate correlation remains the same (or diminish) after partialling out SES. Looking at Colom & Flores-Mendoza (2007) for example, I know it's true. See also Thienpont & Verleye (2004).
MH, thanks for reviewing my pre-submission paper. We are getting ready to submit it, perhaps tomorrow (Saturday).
The one parent must also be Danish citizen. And yes, this legalistic definition causes problems and will increase if exogamy rates go up.
The term "predictor" is just a synonym for "independent variable". It matter not if one uses single OLS regression or correlation, as these are in fact the same. Consequently, the terminology can easily be accommodated.
In case of these longitudinal analyses, predictor also has a temporal meaning to it, in that the IQ was measured a long time before the criteria variables were measured.
Sure, The Bell Curve had the same results (using MR), as did Murray's follow-up papers which I also cite.
A question concerning the first paragraph at page 3. Are you saying that for children who have at least one parent of danish origin, he will be considered danish, no matter what the origin of the other parent is ? If you're interested in trends over time, this way of labeling people can be a confound if the rate of intermarriage is increasing.
The one parent must also be Danish citizen. And yes, this legalistic definition causes problems and will increase if exogamy rates go up.
Also, at page 4, when you cite Strenze for saying that g is a better predictor than parental SES on the child's SES outcome (when he is adult), it depends on what you are thinking. Correlation or regression ? By predictor, you must mean regression. Because it's generally under this context that this word is used. In Strenze's analysis, you have correlation, not regression, so you must replace "better predictor" by "better correlate".
The term "predictor" is just a synonym for "independent variable". It matter not if one uses single OLS regression or correlation, as these are in fact the same. Consequently, the terminology can easily be accommodated.
In case of these longitudinal analyses, predictor also has a temporal meaning to it, in that the IQ was measured a long time before the criteria variables were measured.
I can also add that the information about which of parent SES or IQ is the most important is not the most relevant thing here. What is of more significant finding is whether or not IQ-outcome bivariate correlation remains the same (or diminish) after partialling out SES. Looking at Colom & Flores-Mendoza (2007) for example, I know it's true. See also Thienpont & Verleye (2004).
Sure, The Bell Curve had the same results (using MR), as did Murray's follow-up papers which I also cite.
The term "predictor" is just a synonym for "independent variable". It matter not if one uses single OLS regression or correlation, as these are in fact the same. Consequently, the terminology can easily be accommodated.
In case of these longitudinal analyses, predictor also has a temporal meaning to it, in that the IQ was measured a long time before the criteria variables were measured.
As I said, the term you're using is not clear. By predictor, most people, I think, will wrongly believe you're talking about multiple regression. When people use MR, one usual goal is to estimate which variable is the best predictor. By "correlate" you specify clearly you're not talking about MR. But by saying IQ is a better predictor than parental SES, a lot of readers will be mistaken about what you are referring to.
In longitudinal analysis, predictor has no meaning to me, if it's only a correlation analysis. You can say a correlation between IQ(age9) to SES(age30) shows that IQ predicts later SES. But if you correlate IQ(age40) with SES(age30), you won't talk about IQ as predictor, and yet the correlation can still be very similar.
As I said, if predictor is generally used in the context of MR, you should not employ this word for meaning "correlation analysis".
Sure, The Bell Curve had the same results (using MR), as did Murray's follow-up papers which I also cite.
The Bell Curve mainly uses logistic regression to examine which predictor, between IQ and parental SES, is the best. They don't say if the zero-order correlation between IQ and outcomes is diminished (and to what extent) when SES is partialled out. If anti-IQ people believe IQ is meaningless, they will assume that IQ has no effect on outcome when parental SES is held constant.
In predictive studies (as opposed to other kinds of studies), some variables are chosen as the predictor or independent variables, while others are the ones we want to predict, dependent or criteria variables. The terms do not change just because one switches between OLS regression to correlation, or to some other statistical method. The terms have to do with the chosen role for the variables, not the method.
As for the temporal order, national IQs date back to about 2002 before a large number of immigrants moved to NW Europe. The age heaping data are from 1900, so they are of course earlier. The Islam data from 2010 is not, but extremely highly correlated with the ones from 1990 (r=.99), so the temporal order is fine there too. The int. S factor is from recent data, so the temporal order is not right there. One could without a doubt calculate S factors for older data too that would correlate very highly with the current S factor, so that the temporal order would be correct.
Since the HDI is kind of an S factor measure, and it goes back to 1980, one can use that. http://hdr.undp.org/en/content/table-2-human-development-index-trends-1980-2013
The correlation of HDI 1980 to 2013 is .95. 1980 was the year immigration to Denmark started, so if one can used the present day HDI data, one could have well predicted which groups who would do well.
Murray's follow-up paper did in fact examine the regression weights of IQ when parental SES was controlled. They did this by using the sibling control method, which controls for any kind of shared environment. He showed that the regression found from this method was nearly identical with the one found from MR. He did not, as you say, compare the zero-order with the SES partialed or sibling controlled weight (IIRC).
As for the temporal order, national IQs date back to about 2002 before a large number of immigrants moved to NW Europe. The age heaping data are from 1900, so they are of course earlier. The Islam data from 2010 is not, but extremely highly correlated with the ones from 1990 (r=.99), so the temporal order is fine there too. The int. S factor is from recent data, so the temporal order is not right there. One could without a doubt calculate S factors for older data too that would correlate very highly with the current S factor, so that the temporal order would be correct.
Since the HDI is kind of an S factor measure, and it goes back to 1980, one can use that. http://hdr.undp.org/en/content/table-2-human-development-index-trends-1980-2013
The correlation of HDI 1980 to 2013 is .95. 1980 was the year immigration to Denmark started, so if one can used the present day HDI data, one could have well predicted which groups who would do well.
Murray's follow-up paper did in fact examine the regression weights of IQ when parental SES was controlled. They did this by using the sibling control method, which controls for any kind of shared environment. He showed that the regression found from this method was nearly identical with the one found from MR. He did not, as you say, compare the zero-order with the SES partialed or sibling controlled weight (IIRC).
You don't answer my objection here. Ask what people think when they read "IQ is a better predictor than parental SES". Most of the time, they say MR.
From Wikipedia :
The problem here is that in correlational analysis, there is no independent/dependent variables. But they are considered as such when you use stuff like MR or path analysis.
Usually, predictors or regressors are the variables being manipulated, either by entering other predictors (to be held constant) or manipulated quasi-experimentally through the use of 2-stage least squares (2SLS), a technique that is claimed to be able to circumvent the problem of reverse causality.
As you know, there is no manipulation in the variables used in bivariate correlation. Even if the date of your variable are different, it's no good to use the term "predictor". It's even worse when you use "predictor" just after "better".
From Wikipedia :
An independent variable is also known as a "predictor variable", "regressor", "controlled variable", "manipulated variable", "explanatory variable", "exposure variable" (see reliability theory), "risk factor" (see medical statistics), "feature" (in machine learning and pattern recognition) or an "input variable."[3][4]
The problem here is that in correlational analysis, there is no independent/dependent variables. But they are considered as such when you use stuff like MR or path analysis.
Usually, predictors or regressors are the variables being manipulated, either by entering other predictors (to be held constant) or manipulated quasi-experimentally through the use of 2-stage least squares (2SLS), a technique that is claimed to be able to circumvent the problem of reverse causality.
As you know, there is no manipulation in the variables used in bivariate correlation. Even if the date of your variable are different, it's no good to use the term "predictor". It's even worse when you use "predictor" just after "better".
See here: http://www.openpsych.net/forum/showthread.php?tid=184
How would you like us to phrase it then?
How would you like us to phrase it then?
I think you should write "better correlate" if it's a correlation analysis. By saying this, you make clear that you are talking about correlations, and people will never get confused about if it's correlation or regression.
© Copyright 2024 OpenPsych - All Rights Reserved