#effect size

Subgroup analysis is ongoing.

Effect size is a fundamental concept in statistics that quantifies the magnitude of a phenomenon, relationship, or difference observed in data. Unlike p-values, which indicate whether an observed effect is unlikely to have occurred under a specified null hypothesis, effect sizes convey how large or practically meaningful that effect is. This distinction has become increasingly emphasized in…

Imagine our population data is the red curve and our sample data is the blue and the post will help you make sense of it. We’ve been talking statistics for the past few days and today we’re talking effect size. The short explanation is effect size is the difference between two conditions! The bigger the effect size, the easier it is to tell the two conditions apart, easy… right? There’s a lot…

Yesterday, I had the pleasure of reading an interesting GWAS of Lewy body dementia (LBD) published in Nature Genetics. LBD shares pathological and clinical features with Alzheimer’s disease (AD) and Parkinson’s dementia (PD). So, I was curious to know how the GWAS results of LBD compare to that of AD and PD. As expected, the APOE locus towered over all other loci.  The effect size of the APOE locus was OR-2.46. This is much smaller than the APOE’s effect size for AD. For e.g. this study reports OR-4.6 for AD in heterozygous APOE-e4 carriers and OR-25.4 in homozygous APOE-e4 carriers. So, although APO-E locus seems to account for the largest number of LBD cases (smallest P value and common allele frequency) in this study, its penetrance is markedly lower compared to AD.  

Whenever I read a GWAS, the first thing I look for is which locus has the largest effect size. This cannot be identified using a Manhattan plot, as P values do not reflect the effect sizes. The summary statistics table in the paper showed that a chromosome 1 locus had the largest effect size. At this locus, the index variant is a missense variant in gene GBA, which codes for a lysosomal enzyme glucocerebrosidase. (This is a known locus. A high penetrance of up to ~33% was reported for GBA mutations in LBD in Ashkenazi Jews.) Despite having the largest effect size, this locus was not the one with the smallest P value. Because, this variant is relatively rare and so its P value was not as smaller as APOE’s. This is something that I often emphasise in twitter. A highly penetrant rare variant can have less significant P value than a moderately penetrant common variant. This doesn’t mean that the rare variant is less important than the common variant. It’s in fact the opposite. Variants with larger effect sizes point us to the core disease pathways. In the current study, for e.g., the GBA locus points us straight to the core pathology of LBD: lysosomal dysfunction.

Another perfect example for the effect size vs P value debate is the ACAN locus identified in the 2010 GWAS of height published by the GIANT consortium. I’ve written a twitter thread on this. This locus had the largest effect size, but did not tower tall enough to grab the authors’ attention (a word search for “ACAN” will reveal zero results in this paper’s main text.) The authors wouldn’t had any idea, I believe, that 10 years later, two independent research teams (this and this) will show that the ACAN locus’s large effect association was driven by a variable number tandem repeat (VNTR) within ACAN. The effect size of this VNTR is astonishingly large, in fact, one of the largest ever reported for height. Also, the gene ACAN codes for a cartilage protein thereby falling in the core biological pathway of height. Fascinating, isn’t it ?

Whenever I read a paper, I never miss an opportunity to contemplate on the relationship between sample size, effect size, allele frequency, and phenotypic heterogeneity (the last one is indeed an important factor, but often gets ignored). I see the relationship between these factors analogous to the relationship between aperture size, shutter speed, focal length and ISO in photography (yes, I am an amateur photographer).

Today I glanced through a GWAS paper where the authors report a genetic association (OR-0.54) for nontuberculous lung disease caused by Mycobacterium avium complex (MAC) using a small sample size of few hundred cases and few hundred controls. And they managed to replicate the association and strikingly reproduce the effect size. I am not sure if this finding is a true biological finding (it seems to be). But it made me think how lucky were the authors to identify an association and then replicate it in such a small sample size. If not for this hit, the authors might not have been successful in publishing it in European Respiratory Journal (IF-12.3).

Effect Size and Statistical Significance

Statistical significance is important but not only the most important consideration in evaluating the results. Because, statistical significance tells only the likelihood (probability) that the observed results are due to change alone. It is important to consider the effect size when you obtain statistical significant results.

Effect size is a quantitative measure of some phenomenon like

Co…

This School Year, Prioritize Teacher Clarity to Maximize Student Learning

I heard an awesome quote today… “Clarity is the antidote to anxiety.” I can truly attest to the accuracy of this quote. We all want clarity in our lives. Guess what? Students really want clarity as well, especially in school and in class.

In regards to John Hattie’s influences and effect sizes, you can’t focus on any other influences or effects without first addressing teacher clarity. You can’t…