Decoding The Gurus
patreon
Decoding Academia 19: The Smell Test (Gelstein et al. 2011)
🕑 Added 2023-05-31 02:52:14 +0000 UTCHey everyone,
A free-for-everyone decoding episode in which we look at a pre-replication crisis paper that suggests smelling sad tears makes men less sexually aroused. Bit of an odd paper but also recently promoted by Huberman. We cover some warning flags when it comes to assessing papers, why neuroimaging studies are sometimes misleading, and also get a little bogged down in a discussion of error bars (sorry!).
Hope you enjoy it. Full decoding episode coming soon!
Audio version coming later today.
Comments
Matt
milk sweat
Emma
Ok more fun facts :-) wombat poo is cube shaped!
Adam Session
Fun one to listen to as a stats minded biologist. Agree with many of the criticisms that you bring up. I will directly counter their claim that the within group variance matters more by saying really it is the comparison of within and between group variance that determines the p-value here, not either one on its own. Also to counter one comment Matt made at the end there (professional evolutionary biologist so I have to): we have actually understood the function and evolution of the appendix for a few decades now. It hosts the commensal populations of your probiotic gut bacteria. Essentially diarrhea clears out everything from your main gut, including the bacteria, and the bacteria from the appendix will colonize faster than anything you eat. People who have their appendix removed have issues recovering from diarrhea specifically and have to be treated with fecal transplants, etc. You can also see this reflected in the amount of bacteria needed by the mammal. No mammalian cell can digest dietary fiber found in many plants, but our gut bacteria hcan. Those mammals that eat grass have a lot of the bacteria that digests cellulose, and their appendix is larger to host more of the cellulose digesting bacteria.
Adam Session
Fun fact, monotremes don't have nipples and lactate through their entire abdominal skin.
Christopher Kavanagh
There was a comment this was responding to…
Desi
'Platypus tears' sound like an ingredient to some kind of wizard potion.
Emma
Ok as an Australian animal lover I can't stop myself saying this. Platypus along with echidna are the only living members of the monotreme order. There is three groups of mammals still living: monotremes, marsupials (most of the other ozzy mammals) and placentals. Monotremes are a bit weird though.
Christopher Kavanagh
What is real is what is out there in the real world! Statistical tests can only tell you about relationships observed in your data. They are just tools and like most tools they can be misused- intentionally or by accident. If you want to claim an effect is real ideally what you want is a robust effect that has been produced by multiple independent labs in high-quality studies (reasonable power, pre-registered, good controls). No magic, just science. You won't always have gold standard studies available though so you just have to modulate your certainty in line with the level of evidence available.
Desi
Wait, so what is real then? Seems statistical significance is neither necessary nor sufficient... Do we consider it real when results are shown to be reproducible some number of times? Or does some secret elite club at Science journal HQ sitting around a circular table in an underground bunker cast a vote to decide?
Christopher Kavanagh
Statistical significance is conventionally labelled as anything less than p = .05. However, even if an effect is real, simply by chance you will sometimes get p values above .05. The chances of this increase if your sample is lower powered because there is more chance for noise to influence the results. At low power, you will even with a real effect get a 'dance' of p values, see this video: https://www.youtube.com/watch?v=5OL1RqHrZQ8. So even if an effect is real, you should anticipate seeing some 'insignificant' results by chance (especially if you have low power). If you have a study with a combination of low power, lots of outcome variables, and almost all key outcomes are significant between p = .05 to p = .01, this is a warning sign because that pattern of results is extremely unlikely. There is another video by Daniel Lakens that might help explain: https://www.youtube.com/watch?v=RVxHlsIw_Do
Demostix
I get all the points about low sample sizes, multiple degrees of researcher freedom etc, but I don't get the "close to significance" argument. I thought significance was below 0.05, so I'd get the argument if they hovered around say 0.04 or something like that. Not defending the paper in any way, I just wanna understand the argument.
