Tao Coll-Martín

University of Granada (@cimcyc | @CanalUGR) - Psychology research - #ADHD #TDAH

[Luis Cásedas - @lcasedas.bsky.social @lcasedas]

Curious about the mind? Happy to announce the Insights in Mind Sciences (INMINDS) interview series! Link at the end of this short thread. 🧵1/5

[Luis Cásedas - @lcasedas.bsky.social @lcasedas]

Our theoretical perspective on the mechanisms underlying mindfulness-induced cognitive change is now out in @NatRevPsych!

[Tao Coll-Martín @tcollma]

@lakens Is it correct to decide to collect more data (while adjusting for alpha) based on post-hoc criteria such as the belief that an effect is likely present, because the initial data showed a meaningful mean effect size and/or a non-significant but low p-value?

[Tao Coll-Martín @tcollma]

@lakens And when you have specific hypos but cannot keep the error rate nominal (because you have multiple comparisons)? This is a very common case where you cannot do severe testing, but you have clear hypos and you can calculate the (inflated) error rates of your claims.

[Tao Coll-Martín @tcollma]

@lakens What would you call this type of research where you control the error rate (e.g. in a correlation matrix) without having concrete hypotheses?

[Tao Coll-Martín @tcollma]

@Fran_gfr Interesante amigo Fran. Imagino que esa mayor variabilidad 'entre' de los estudios online, no se debe tato a diferencias individuales (personas más heterogéneas), sino a diferencias en el setting donde se hace la tarea, que también se mantendrían estables incluso en test–retest

[Tao Coll-Martín @tcollma]

@FBpsy Entonces, para hablar fraudes como los que describes, creo que habría que especificar en la sección de Transparency qué hipótesis/análisis se hicieron antes de recoger/ver los datos y cuáles después (y cuáles se planearon pero no se reportan). Engañar ahí sí sería fraude

[Tao Coll-Martín @tcollma]

@FBpsy Quizá el problema está en que los lectures asumimos que todo es confirmatorio salvo que se especifique lo contrario

[Tao Coll-Martín @tcollma]

@dingding_peng @MaxPrimbs @Flavio_Azevedo_ @EikoFried Perhaps the problem here is that advocates of tiny effect sizes may argue that increasing locus of control can have many small positive impacts on different variables that, in combination, make any variable minimally affecting locus of control matter 😵

[Tao Coll-Martín @tcollma]

This work has been carried out using the ANTI-Vea-UGR, the public platform of our lab (led by @jlupiane) to measure attention online and free: https://t.co/gLr4Qw2xtd

[Tao Coll-Martín @tcollma]

TLTR: Our replication doesn't support the idea that adult-onset ADHD is neuropsychologically distinct from conventional (childhood-onset) ADHD — at least, in terms of vigilance and from an ADHD dimensional model

[Collabra: Psychology @CollabraOA]

New from Tao Coll-Martín @tcollma, Hugo Carretero-Dios & Juan Lupiáñez: Attention-Deficit/Hyperactivity Disorder Symptoms as a Function of Arousal and Executive Vigilance: Testing Halperin & Schulz’s Neurodevelopmental Model in a Sample of Community Adults https://t.co/VoN6vNG44y

[Tao Coll-Martín @tcollma]

*For power analysis, we based on estimates of the attenuated effect sizes due to the random measurement error (reliability) of the variables involved in each test

[Tao Coll-Martín @tcollma]

@Neurorandom Quizá te interese https://t.co/idFpEOC7jN

[Tao Coll-Martín @tcollma]

@uri_sohn @stephensenn Thanks for the advice 🙏 Indeed, your point is that GAM is better than quadratic, but that in complex models GAM is only achievable by a minority of social scientist :)

[Tao Coll-Martín @tcollma]

@uri_sohn @stephensenn But in the case of complex data structures (as you mentioned), why your suggestion is lm(y~x*z+I(x^2)+I(z^2)) instead of lm(y~x*z*I(x^2)*I(z^2))?

[Tao Coll-Martín @tcollma]

@uri_sohn @stephensenn 2) Would there be a risk of over-fitting the model (especially if there is no reason to think of non-linear relationships)?

[Tao Coll-Martín @tcollma]

@uri_sohn @stephensenn Very good paper, I'll definitely add the quadratic terms to test for interactions in my observational studies. My only two questions are: 1) Why don't you include interaction with quadratic terms (main effects only)?

[Tao Coll-Martín @tcollma]

@lakens @JakeJares Does it make sense for you pure estimation based on descriptives when inferential analyses are severely underpowered? — instead of relying on p-values that would probably lead to a high FDR

[Tao Coll-Martín @tcollma]

@lakens @SpikyBouba Analogously to how in the secondary analysis robustness checks could serve severity assuming a given direction of the prediction would have been taken

[Tao Coll-Martín @tcollma]

@lakens @SpikyBouba My (maybe naive) take is that p values could be informative to other researchers or counterfactual scenarious in which this test would follow from a planned/confirmatory analysis