Lukas Lengersdorff
@lukaslengersdorff.bsky.social
PhD student @University of Vienna. <3 Stats, Modelling, Philosophy of Science
To be fair, good power calculations are also much more difficult than they are made out to be. T-tests, correlations, sure, just use GPower, but as soon as things get more complicated (e.g. LMMs), power calculations become small simulation studies in themselves
February 2, 2025 at 7:43 AM
To be fair, good power calculations are also much more difficult than they are made out to be. T-tests, correlations, sure, just use GPower, but as soon as things get more complicated (e.g. LMMs), power calculations become small simulation studies in themselves
And yeah, the risk for misunderstanding/misrepresentation/oversimplification always remains... but if we only published papers that could not possibly be misrepresented, journals would get pretty thin. I'd hope that review/supervision will act as a corrective to this
February 1, 2025 at 8:54 AM
And yeah, the risk for misunderstanding/misrepresentation/oversimplification always remains... but if we only published papers that could not possibly be misrepresented, journals would get pretty thin. I'd hope that review/supervision will act as a corrective to this
Thank you for the kind words! I do hope though that you read it in the parking lot *before* you picked up your kid(s) 😄
February 1, 2025 at 8:51 AM
Thank you for the kind words! I do hope though that you read it in the parking lot *before* you picked up your kid(s) 😄
And all of that is independent of the (probably much more important) other message of the paper: if your real issue with the result is that you cant imagine it to have been obtained without questionable research practices, than that's the much bigger problem than the power
January 31, 2025 at 4:48 PM
And all of that is independent of the (probably much more important) other message of the paper: if your real issue with the result is that you cant imagine it to have been obtained without questionable research practices, than that's the much bigger problem than the power
In my (maybe naive) view, less-than-ideal power is rarely a deliberate choice, but rather the consequence of limited resources, and maybe honest mistakes in study planning. That's not great - but if sig. results are produced in an honest way anyway, I think it'd be good to assess them at face value
January 31, 2025 at 4:43 PM
In my (maybe naive) view, less-than-ideal power is rarely a deliberate choice, but rather the consequence of limited resources, and maybe honest mistakes in study planning. That's not great - but if sig. results are produced in an honest way anyway, I think it'd be good to assess them at face value
I do find it interesting, however, that a lot of malicious intent gets assigned to low power. Do we really expect that there are many people thinking "muahaha, I shall now run a LOW POWER study to advance my career/hurt science/publish non-replicable results"?
January 31, 2025 at 4:39 PM
I do find it interesting, however, that a lot of malicious intent gets assigned to low power. Do we really expect that there are many people thinking "muahaha, I shall now run a LOW POWER study to advance my career/hurt science/publish non-replicable results"?
Re abuse: that worry might have crossed our minds once or twice... or a million times. Which is why we added MANY reminders to the paper that "deliberately" conducting low power studies is (a) bad for science (b) bad for yourself (why would you want to not have a good chance to detect your effects?)
January 31, 2025 at 4:33 PM
Re abuse: that worry might have crossed our minds once or twice... or a million times. Which is why we added MANY reminders to the paper that "deliberately" conducting low power studies is (a) bad for science (b) bad for yourself (why would you want to not have a good chance to detect your effects?)
(first author here)
Re overestimation: imho only in interaction with publication bias. That would def. have been an interesting point to discuss. But we focussed on the implications of low power in hypothesis testing (yes/no decision). Adding estimation would probably have made the paper explode 🤯
Re overestimation: imho only in interaction with publication bias. That would def. have been an interesting point to discuss. But we focussed on the implications of low power in hypothesis testing (yes/no decision). Adding estimation would probably have made the paper explode 🤯
January 31, 2025 at 4:29 PM
(first author here)
Re overestimation: imho only in interaction with publication bias. That would def. have been an interesting point to discuss. But we focussed on the implications of low power in hypothesis testing (yes/no decision). Adding estimation would probably have made the paper explode 🤯
Re overestimation: imho only in interaction with publication bias. That would def. have been an interesting point to discuss. But we focussed on the implications of low power in hypothesis testing (yes/no decision). Adding estimation would probably have made the paper explode 🤯