Championship Subdivision Football | FCS Football | Stadiums | Blogs | Forums

Had a medical procedure so I am home with time to play. I took the average score of the 2013 US Department of Education National Assessment of Educational Progress (NAEP) math test for each of the 50 States. I went to the Census Bureau web page. For the impact of Socioeconomic Status I took the estimate percent of people in each State in poverty. For "educational attainment" I took the estimate of the percent of people in each State who have a bachelors degree. For race I took the sum of the estimates of the percentages of non hispanic Whites and Asians in each State. I looked at the combined percentage of Asian and Non Hispanic Whites because those are the two racial groups that scored above the overall national average on the test. So I had the variables Socioeconomic Status, Educational Attainment, and Race.

I ran all possible models using those variables. So I had the models:

Socioeconomic Status
Educational Attainment
Race
Socioeconomic Status + Educational Attainment
Socioeconomic Status + Race
Educational Attainment + Race
Socioeconomic Status + Educational Attainment + Race

What happened? The model that accounted for the greatest extent of variation in test scores AND which had all variables demonstrating "significant" effects is "Educational Attainment + Race." Simply taking the percentage of persons in each State who held a bachelors degree and the percentage of persons in each State who are either Non Hispanic White or Asian into account accounted for 69% of the variation in State average scores. That is a striking illustration of the point that most of the variation one sees between states in standardized test scores can reasonably be attributed to demographics.

BTW, Educational Attainment, Race, AND Socioeconomic Status accounts for a little more of the variation. If you put all three of those in the model 71% of the variation is accounted for. But the problem is that, in that model, Socioeconomic Status is not "significant."

What I did is not the best way to do it. The best way to do it would be to have all the data on each individual student who took the test. But the results of what I did are consistent with the belief that you can come pretty close to telling how States are going to rank in terms of standardized test scores just by going to the Census Bureau web site and looking at the demographics of each State. How States rank is much more a matter of the demographics each State education system has to deal with than it is a matter of how good different State educations systems are at doing the job.

I would think you might have multicollinearity issues with race and SES.

I think you'd enjoy this book

http://www.amazon.com/Passing-Great-Rac ... B008KNW1Q0" onclick="window.open(this.href);return false;

So in other words, you suck at educating poor black people? Shocking!

Pwns wrote:I would think you might have multicollinearity issues with race and SES.

No. Remember, SES fell out of the model. To the extent that there may have been multiollinearity issues involving both variables capturing much of the same thing one of them fell out. And it was the "SES" variable.

Besides, by the reference I use, multicollinearity would not have been unacceptable even if race had stayed in the model. I looked at that too.

kalm wrote:So in other words, you suck at educating poor black people? Shocking!

No, you're getting it opposite. If you look at Louisiana, for instance, poor Blacks in that State scored about the same as poor Blacks nationally. To the nearest whole point the national average is 258 and so is the Louisiana average. Slightly less if you go to decimal points at 257.57 to 258.30, but that's within the standard error.

And if you include parental education level, the most disadvantaged group in Louisiana...poor Blacks with parents that did not graduate high school...scored slightly higher on average that that group did nationally. It was 253.94 to 252.83. Again within a standard error.

The problem for Louisiana is that there are a whole lot MORE people in those disadvantaged groups proportionately in that State than there are in those disadvantaged groups nationally.