Talk:Scarr–Rowe effect

Interpreting 2018 and 2021 studies

I would invite the IP user most recently editing as 2600:1700:77C0:EFF0:8108:9DB0:C08:C5F7 to discuss their preferred edits here rather than edit warring, and to take a look at the policies I've mentioned in my revert summaries, especially WP:OR. Thanks, Generalrelative (talk) 15:17, 6 February 2022 (UTC)

For reference, the two studies in question here are [1] and [2]. Generalrelative (talk) 03:04, 7 February 2022 (UTC)

Thank you for the offer to explain myself. I do not think it is original research to say β = .08 is equal to less than 1 percent of variance explained, or atleast a small/weak effect. I am open to being corrected if I mischaracterized anything.2600:1700:77C0:EFF0:2DCD:AFF:3A38:E06D (talk) 00:58, 7 February 2022 (UTC)

Thanks for engaging with me. The test for original research is quite straightforward. Can you point to somewhere in either of the studies where this is stated explicitly? If not, then it is original research. I will be happy to be proven wrong but I did not see it. Generalrelative (talk) 01:06, 7 February 2022 (UTC)

As a side note: I am no expert here, but it is not at all clear to me that "β = .08 is equal to less than 1 percent of variance explained" as you say. The source you linked to in one of your edit summaries (Table 2 in this paper: [3]) does not show a linear relationship between β and variance explained. And indeed, it seems counterintuitive for the 2018 study to show less than 1 percent variance explained and yet describe their findings as "robust" and "highly statistically significant". Regardless, this latter point is tangential to the main issue. The question of whether the addition is original research is addressed above, and that should be the focus of this discussion. Generalrelative (talk) 01:11, 7 February 2022 (UTC)

A effect size can be statistically significant/exist regardless of model used and still be small, see </ref>Solla, F., Tran, A., Bertoncelli, D., Musoff, C., & Bertoncelli, C. M. (2018). Why a P-Value is Not Enough. Clinical spine surgery, 31(9), 385–388. https://doi.org/10.1097/BSD.0000000000000695</ref>. I have provided sourcing (not my own subjective view) showing a effect of β=.10 translates to .01, or 1 percent of the variance, and since the β values found in both papers are less than β=.10, it isnt inaccurate to say the effect sizes found explain less than 1 percent of the variance. If you or anyone else can show what if anything I have mischaracterized/misintepreted in this response, please do so. — Preceding unsigned comment added by 2600:1700:77C0:EFF0:2DCD:AFF:3A38:E06D (talk • contribs)

Thanks for the additional ref. I'll look into it, but for the time being I hope my explanation of original research above is helpful. If the authors do not go out of their way to state that the effect size is small then neither will Wikipedia. Generalrelative (talk) 01:43, 7 February 2022 (UTC)

Having looked over both studies again just now, I have to say that I find them to be quite sanguine about having replicated the Scarr-Rowe effect, so our presentation of these studies needs to reflect that. Doing otherwise (i.e. by highlighting a detail not emphasized by the authors) would violate the principle of WP:DUE weight. Generalrelative (talk) 01:55, 7 February 2022 (UTC)

Is there any way even for just a small note at the end of the page that the effects found were small? — Preceding unsigned comment added by 2600:1700:77C0:EFF0:CB3:30AE:8C43:4110 (talk) 02:33, 7 February 2022 (UTC)

It isn't up to me. WP:OR is a core policy. If you can show where these studies state explicitly that the effect size was small then we can discuss whether inclusion would be WP:DUE. Generalrelative (talk) 03:07, 7 February 2022 (UTC)

The IP is right that effect size is different from statistical significance. Parapsychology, for example, typically has tiny effect sizes with high significance, likely caused by occasional information leaks.

But I cannot see any evidence that the effect size is small in this case. To the contrary, [4] says, Turkheimer et al. (Reference Turkheimer, Haley, Waldron, D'Onofrio and Gottesman2003) reported one of the largest Scarr–Rowe effects, finding that among those with the lowest SES, the heritability of IQ was close to zero.

I don't know where the number .08 came from, but "β = .08" means that the probability of such a result occurring by chance is 8%. There is no connection to effect size. --Hob Gadling (talk) 07:31, 7 February 2022 (UTC)

This “B” type character can refer to either the beta level or the beta coefficient (please, correct me if I’m wrong)“The results are similar to those produced using a more conventional two-way interaction model, with IQ predicting the EA3 × log of parental SES interaction after the main effects; however, CPEM allows for greater model degrees of freedom, thus is better powered to detect the effect when it is small in magnitude (CPEM β = 0.05, p = 6.69×10−5 vs. two-way interaction β = 0.02, pone-tailed = .045, in both models log parental SES is used).” So looks like the effect is indeed small, and smaller than β = .08 — Preceding unsigned comment added by 2600:1700:77C0:EFF0:F584:F321:9F1F:1D68 (talk) 11:38, 7 February 2022 (UTC)

β can have several meanings in different contexts - see Beta (disambiguation)#Mathematics, finance, and statistics - but I don't see why β should be anything other than the beta level, when it always appears together with its close cousins, the p-value p and the degrees of freedom df. --Hob Gadling (talk) 13:02, 7 February 2022 (UTC)

Notice how in the excerpt quoted from the abstract the effect (not statistical significance) is shown to be β = 0.05/β = 0.02, it is even clarified they are using a model that is better at detecting small effects. — Preceding unsigned comment added by 2600:1700:77C0:EFF0:24A7:C001:8B75:B701 (talk) 20:51, 7 February 2022 (UTC)

I don't know what you mean by the excerpt quoted from the abstract, but searching both pages for ".05" led me to this passage: Example text I guess you mean to interpret "slightly lower magnitude effect size (β = 0.02 vs. 0.05)" as: β is the magnitude. That interpretation is wrong, as you can see from the fact that 0.02 is not "slightly lower" than 0.05, it is less than half of it! I don't know why the β values are given at this place, but your conclusion that β is the effect size does not hold water.

Can we please stop this? β is not an effect size. Your deductions are WP:OR anyway, even if they were correct. Find a source that explicitly says it or leave it alone. --Hob Gadling (talk) 07:13, 8 February 2022 (UTC)

Oh, another thing. If β had another meaning at this place than everywhere else in the article, where is clearly the beta level, it would mean that the authors made the absolute rookie mistake of using the same symbol for two different things in the same article. If they did that, then the whole source would be crap and therefore unusable. --Hob Gadling (talk) 07:22, 8 February 2022 (UTC)

On page 5 of the paper“ The reduced model degrees of freedom in the interaction model (6,253 vs. 6,255), ****coupled with the slightly lower magnitude effect size (β = 0.02 vs. 0.05),**** led to a non-significant value for the EA3 × parental SES interaction term. Despite this, given that the results of the CPEM analysis and the prior meta-analysis of U.S. stud- ies based on the use of twins and siblings permit a direc- tional prediction of the interaction effect to be made, the use of one-tailed significance is justified in this instance (Kimmel, 1957), which yields a significant result (p = .045).” — Preceding unsigned comment added by 2603:6010:6046:5200:0:0:0:1 (talk) 22:30, 9 February 2022 (UTC)

So, can it now be agreed that the largest effect found was β = .05, and that this is a small effect size? — Preceding unsigned comment added by 2600:1700:77C0:EFF0:8DED:32FB:8A99:B9B6 (talk) 23:46, 20 February 2022 (UTC)

No. Generalrelative (talk) 23:50, 20 February 2022 (UTC)

What did I say that you take issue with? — Preceding unsigned comment added by 2600:1700:77C0:EFF0:8DED:32FB:8A99:B9B6 (talk) 00:28, 21 February 2022 (UTC)

Hob Gadling has explained it perfectly well just above. I suggest you go back and reread that answer. Generalrelative (talk) 00:37, 21 February 2022 (UTC)

Gadling in his response did not refute that the cited excerpt was talking about effect sizes, "That interpretation is wrong, as you can see from the fact that 0.02 is not "slightly lower" than 0.05, it is less than half of it! I don't know why the β values are given at this place, but your conclusion that β is the effect size does not hold water" the only criticism was that .05 to .02 is a difference that is misleading to characterize as "slightly lower". Saying that this peer reviewed paper would be "unusable" because it uses the same character for significance or effect does not hold water when you consider the paper specifies whether significance or effect is being referred to. Also, (Turkenheimer et al 2003) does not have anything to do with the findings of (Woodley et al 2018.) — Preceding unsigned comment added by 2600:1700:77C0:EFF0:8DED:32FB:8A99:B9B6 (talk) 01:14, 21 February 2022 (UTC)

I suggest at this point that you stop wasting your time and that of others. Whether or not β refers to the beta levels in these studies (which seems prima facie to be the obvious meaning), we will not be stating in the article that the effect size is small unless the cited sources say so explicitly. That is, with words. Doing otherwise would be WP:UNDUE emphasis at best, and very likely WP:OR. So even if there were some chance you might persuade others here that β refers to the effect size, which seems unlikely at this point, it would still be idle chatter. And article talk pages are not the place for that. As far as I'm concerned, therefore, –– until someone else shows up with knowledge on the matter –– this conversation is over. Generalrelative (talk) 01:33, 21 February 2022 (UTC)

Its prima facie that when values are given after the effect is singled out, the values are indeed the effect given. IF you want explicit wording, then read page 2 of (Woodley et al 2021) where this is said "In the study of Woodley of Menie et al. (2018), two methods

were used to estimate the interaction between a cognitive PGS (designated EA3 in Lee et al., 2018) and parental SES. First, the Continuous Parameter Estimation Model was used to estimate the interaction between the two directly, without estimating the main effects (β = 0.08, SE = .01, p = 4.71 × 10−10). Second, the effect was also detected using a conventional two-way interaction model in which the interaction between the two was estimated hierarchically, after first estimating the main effects of EA3 and (log-transformed) parental SES on IQ (β = 0.02, SE = .01, pone-tailed = .0451 ). The positive signs on these coefficients indicate that as the level of parental SES increases, so too does the covariance between EA3 and IQ. In turn, this increase in covariance can be said to capture the increased expressivity of EA3 on IQ as a function of rising parental SES." — Preceding unsigned comment added by 2600:1700:77C0:EFF0:8DED:32FB:8A99:B9B6 (talk) 01:57, 21 February 2022 (UTC)

This quote does not say the effect size is small either. You have nothing. Go away. --Hob Gadling (talk) 08:59, 21 February 2022 (UTC)

It is a small effect size regardless, as β^∗<0.2 is considered a weak, 0.2<β^∗<0.5 moderate, and β^∗>0.5 strong effect. Source: Acock, A. C. (2014). A Gentle Introduction to Stata (4th ed.). Texas: Stata Press. (Page 272). Considering that the effect found is less than β=.2, it is less than weak. Its the same reason you can say β=.02 is less than half of β=.05 even though its not explicitly said in the paper. — Preceding unsigned comment added by 2600:1700:77C0:EFF0:8472:D835:DB96:5F4A (talk) 21:53, 21 February 2022 (UTC)

Small/large and weak/strong do not have anything to do with each other. β is about weak/strong. Read WP:CIR, then stop this. You have no clue, and you cannot contribute to improving the article. --Hob Gadling (talk) 07:35, 22 February 2022 (UTC)

So then it is not inaccurate to say Woodley et al 2018 found a weak effect, are we in agreement? If I do not have sufficient competence because I mixed up small/large, weak/strong, what does citing (Turkenheimer et al 2003) to make claim about the effect of (Woodley et al 2018) say about competence of the poster? — Preceding unsigned comment added by 2600:1700:77C0:EFF0:8DE8:461E:2941:8E05 (talk) 15:37, 22 February 2022 (UTC)

You either did not understand Hob Gadling's point in mentioning Turkheimer or are pretending not to. The authors are clearly comfortable saying in plain English when an effect size is large and yet do not say that their effect size is small. At this point I strongly suspect that you are simply trolling, or perhaps attempting to push some obscure ideological POV. In either case, the thing to do going forward is to WP:DENY recognition. My silence should at no point be taken for consent. Generalrelative (talk) 15:54, 22 February 2022 (UTC)

It is already established that the effect found was β=.02, and that this is a less than weak effect size, Gadling has not refuted this, choosing to criticize my use of the word "small" instead of "weak" to describe the effect. — Preceding unsigned comment added by 2600:1700:77C0:EFF0:C51:1B57:B005:7236 (talk) 20:02, 22 February 2022 (UTC)

"Weak effect size" does not make sense. "Weak" in this context means "the uncertainty is high" or "the error bars are large". An effect can be large with large error bars (large, weak), large with small error bars (large, strong), small with large error bars (small, weak), or small with small error bars (small, strong). --Hob Gadling (talk) 06:51, 23 February 2022 (UTC)

Hur & Bates 2019

This reference was removed for reason of ̈Study does not mention the Scarr–Rowe effect". Turkenheimer et al 2003 does not mention the Scarr-Rowe Effect by name either yet is still cited by the article. Because both papers investigate the impact of SES on genetic influence in IQ. — Preceding unsigned comment added by 2600:1700:77C0:EFF0:7C61:6C22:B238:AAAC (talk) 23:32, 22 February 2022 (UTC)

References