Correction of Flynn Effect

Correction of Flynn Effect                                                                                       -------WANG XI

CONTENTS:

Abstract

  Keywords

Introduction

Methods

Results

Discussions

Conclusion

References

Abstract:

The “Flynn effect” concerns with IQ increase over time and generation, which is a social widely accept theory.

Keywords:

Flynn effect, IQ test, generation, special education, high stakes

Introduction:

In our daily life, we always say that human is getting cleverer, which means this theory assume IQ is an increasing item over time.

We will see how it is by Flynn effect. The “Flynn effect” refers to the observed rise over time in standardized intelligence test scores, it shows during 1932 to 1978, intelligence scores increased 0.3 per year.

But can we claim we are smarter than previous person? We still need to discuss some other respects: brain and intelligence, effects of gene and environments.

Flynn effect: At first, prof. Flynn studied this subject in order to object the idea --–Black people is low inherently intelligent. Flynn’s study claim that IQ rate increase at 0.31% per year during 1932-1978, which is supported by 7431 participants.

We assume there are 2 mean drivers for this case, Gene and environments.

For gene, according to Mingroni’s study in 2007, IQ gains are the result of increasingly random mating, but it is not very useful when we check the data about 1932-1978.

For environment, prenatal environment Improvements and socioenvironmental improvements do affect on Flynn effect.

There is one huge problem, which factors lead the increase of Flynn effect’s result.

Methods:

First of all, we set up the homogenous sample: We limited our primary investigation to comparisons between tests with greater than five years between norming periods. The reason is that we can reduce the error and magnify difference between data effectively.

Secondly, we use 27 tests to analysis different goals.

Then, there is a Moderator analysis. We divided participants into an old set and a modern set.

Next, we did such Statistical Methods, for example, we concentrated on Comprehensive Meta Analysis software, and especially focused on an effect size for every study; we changed the size weighting to find out the changes.

Finally, we use statistical methods to get the Effect size metric (M = 100, SD
= 15 or 16), the effect size weight, credibility intervals and selection of random effects model. Then we can get heterogeneity in effect sizes by using this formula:

I^2=(Q-df)/Q. Finally, we get one publication bias from all these data and models.

Results:

According to the methods, the mean effect over 285 total studies (n = 14,031) in the random effects model was 0.231 IQ points per year, 95% CI [0.20, 0.26], z = 14.10, p < .0001, with a confidence interval and p-value indicating that the Flynn effect is different from zero1. The effects were significantly heterogeneous, (Q(284) = 4710, p < .0001). The estimated I2, or proportion of the total variance due to true study variance, was I2 = 0.94. The Tau, or estimated standard deviation of the true effects, was τ = 0.25, resulting in a credibility interval of −0.26 to +0.72. Eighty-two percent of the distribution of true effects was above zero.

During the model test, we expected some useless figures, we got results that there were still 53 effects of 0.293 points in every year, the 0.95 confidence interval is [0.23,0.35], so we claim the regression of effect on ability is not significant at all. For age, the regression of effect was not significant any longer. For sample type, we consider type of sample was nonsignificant in case of random effects analysis. For order effects, we also thought it was a nonsignificant factor.

When we turn our focus to the meta-analysis, we can see that when we compare the given correlation between the two tests (pervious study), we conclude all significance tests and heterogeneity tests achieved same conclusions. We got 93 more effects with norming gaps of 5 years. According to these 378 studies we claim round 0.86 of the distribution of true effects was greater than 0.