It's been theorized for a long time that elephants are able to communicate via conducting sound through their feet, which seem to be specialized to pick up very low-frequency sound waves. It's...
It's been theorized for a long time that elephants are able to communicate via conducting sound through their feet, which seem to be specialized to pick up very low-frequency sound waves. It's thought that elephants will even choose different kinds of terrain to stand on to be able to listen or rumble better.
For those interested in the mechanics of this, there is some college-level biology stuff here with one image in particular that I was hoping would be a bit more detailed, but at least the rest of the document seems to explain it well.
There are also some YouTube videos about this phenomenon (one TEDx by Dr. O'Connell from 2013) that you can watch from this article.
Elephants are really cool and social creatures and we're constantly learning more about them and proving hypotheses!
npr 469 doesn't sound like a huge sample of calls, and how many of those have "name calls", and then 27% of the time guessed correctly is...how many instances? Does it make sense for me to feel as...
What they found is that their model was able to identify the correct elephant recipient of the call 27.5% of the time, which is much better than it performed during a control analysis that fed it random data, says Pardo.
469 doesn't sound like a huge sample of calls, and how many of those have "name calls", and then 27% of the time guessed correctly is...how many instances?
About 470 distinct calls were captured from 101 unique callers corresponding to 117 unique receivers in Samburu National Reserve and Amboseli National Park. another article
Does it make sense for me to feel as skeptical as I am? What if the enthusiastic response is due to the call meaning something like "come here" and "oooh check this out if you're part of my herd " and "food for my babies" compared to non interesting things like "food but only if you're my baby". Or if the response is due to how much they miss the caller? Or even how muffled the recording is, making it a sort of, oh that's my mum but I can't hear what she's saying I better get closer.
They haven't isolated the "names" yet it sounds like. The next step would be if the researchers could reproduce just the name and use that to attract specific elephants? I don't know why I'm being so picky about this and not as excited about it as when I learned about bees wanting to play with balls
Disclaimer: I only have an entry level statistics and a biostatistics course for my background. All statistical data can be presented in such a way to support almost any conclusion. But I don't...
Disclaimer: I only have an entry level statistics and a biostatistics course for my background. All statistical data can be presented in such a way to support almost any conclusion. But I don't think that's happening here.
EDIT: Notcoffeetable and C-Cab's responses below are much more useful as to the discussion of the sample size aspect. My statement about 20 samples isn't really relevant here, and there's a lot more nuance to sample size than just saying there's a minimum number.
Generally, you need a sample size of at least 20 to have enough data for statistically significant predictions to be made, so 470 (ANOTHER EDIT: It's actually 101 unique elephants, 469 separate recorded calls) isn't too bad of a sample size for biology research. I just perused the paper itself very briefly, and they discuss some of your skepticisms in the Discussion/Conclusion section. Additionally, their control model consisted of ten thousand separate models that each contained nothing but random acoustic permutations, and none of those 10,000 models could identify receivers based on their calls more than 10% of the time. Their own model could identify 27.5% of receiver identities, and according to their statistical analysis in the paper, there is less than 0.01% chance of that happening due to a random coincidence (P<0.0001). I think it seems pretty likely that elephants are actually communicating with each other specifically in some way from this paper's data. Quoted below is also a relevant paragraph from the discussion section to your comment:
We also controlled for behavioural context and recording date in the proximity score analysis, ensuring that receiver specificity was not an artefact of context-related cues or autocorrelation among calls from the same day. The results did not change when two individuals that accounted for a disproportionate number of calls in the dataset (M6 and M6.99) were excluded, indicating that our results were not driven by a few highly influential individuals (Supplementary Information). Most importantly, elephants responded more strongly to playback of calls addressed to them than to playback of calls from the same caller addressed to a different receiver, indicating that the calls contained receiver-specific information that was salient to the elephants. The difference in response to test and control trials was often pronounced. For example, subject R26 vocalized eight times and approached the speaker in response to the test playback but vocalized only once and did not approach the speaker in response to the control playback. Only one subject exhibited an unambiguously stronger response to the control playback than to the test playback. These results are particularly notable in that we could not be certain that all playback stimuli contained vocal labels.
It's important to keep in mind that there is no standard number of samples you need that applies to every study. Estimated effect size is super important for determining the statistical power of...
It's important to keep in mind that there is no standard number of samples you need that applies to every study. Estimated effect size is super important for determining the statistical power of any given study design, such that smaller effect sizes will need more samples to identify statistical differences. In my experience, a lot of physiological experiments can get away with fewer samples as a result of this, but behavior data needs a bit more as the variance is much higher.
You're right, thanks for pointing this out. I've edited my other comments to try and clarify and point towards you and Notcoffeetable's comments regarding sample size.
You're right, thanks for pointing this out. I've edited my other comments to try and clarify and point towards you and Notcoffeetable's comments regarding sample size.
Going to need a citation on that. From what I know, good sample sizes range from 100 to 10% of the population size, but it ultimately depends on a lot of nuances of the study design.
Generally, you need a sample size of at least 20 to have enough data for statistically significant predictions to be made,
Going to need a citation on that. From what I know, good sample sizes range from 100 to 10% of the population size, but it ultimately depends on a lot of nuances of the study design.
Generally the number given in basic stats classes is about 40, some fields will accept 20 (in particular exercise science comes to mind where the interventions are invasive enough to make...
Generally the number given in basic stats classes is about 40, some fields will accept 20 (in particular exercise science comes to mind where the interventions are invasive enough to make recruitment difficult.)
I am a mathematician not a statistician but the value 40 gets tossed around because I believe the standard techniques for computing p-values are well behaved. But I think (correct me if I am wrong) a statistician would not be happy with n=40 (much less 20) unless that was a decent portion of the population.
I work in a more statistical role, and generally I don't do much with 500 data points when it comes to ML modeling.
Edit: after a bit of reading, the number I'm seeing tossed around is n=30. And this is more or less a heuristic based on the Central Limit Theorem and its rate of convergence.
Sorry, I should have been more clear and mentioned that 20 is an absolute bare minimum and not an ideal number. ̶W̶i̶t̶h̶ ̶t̶h̶a̶t̶ ̶s̶a̶i̶d̶,̶ ̶a̶t̶ ̶4̶7̶0̶ ̶e̶l̶e̶p̶h̶a̶n̶t̶s̶ ̶w̶e̶ ̶a̶r̶e̶...
Sorry, I should have been more clear and mentioned that 20 is an absolute bare minimum and not an ideal number. ̶W̶i̶t̶h̶ ̶t̶h̶a̶t̶ ̶s̶a̶i̶d̶,̶ ̶a̶t̶ ̶4̶7̶0̶ ̶e̶l̶e̶p̶h̶a̶n̶t̶s̶ ̶w̶e̶ ̶a̶r̶e̶ ̶f̶a̶r̶ ̶b̶e̶y̶o̶n̶d̶ ̶t̶h̶a̶t̶ ̶l̶i̶n̶e̶.̶ EDIT: It's actually 101 unique elephants, 469 separate calls recorded. And when you consider how many resources it must take to track elephants and record what they do acoustically, 101 separate elephants is probably a realistic sample size. But also only having a sample size of 101 is probably a contributing factor to why their model is only identifying 27% of recipients. If they could've sampled more individuals, accuracy might be higher.
Regarding citation, I am trained as a medical lab scientist in the US, and was trained that 20 data points is the minimum needed when you're doing a validation on a new test that you're bringing into your lab, so that's where my number 20 came from. You need 20 data points from the new and old assays to compare (if you have a previous version of the assay.) Ideally, you get more than 20, but that's not how healthcare operates in the real world oftentimes 😆 Additionally, if your lab identifies the reference ranges for your own population (for example, what is the reference range for glucose for our patient population?), I was trained that you need at least 20 samples when identifying that. I understand this isn't really analogous to measuring unknowns about elephants, but I'm just explaining where that number came from.
Quick googling found this article, which supports having 20 samples for patient reference ranges towards the end of it, although additionally 95% of those samples have to line up with the published reference range for the analyte being measured: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10349316/
So yeah, 40 is probably a better minimum number lol
Worth noting: it's 101 elephants. 470 unique calls. But as a linguist, I don't think it's a bad sample size, especially to just explore the idea of whether these are distinct naming calls.
Worth noting: it's 101 elephants. 470 unique calls. But as a linguist, I don't think it's a bad sample size, especially to just explore the idea of whether these are distinct naming calls.
One important thing to keep in mind here is that sample size should be determined by the statistical power of your study design which is going to be determined by the magnitude of the effect size...
One important thing to keep in mind here is that sample size should be determined by the statistical power of your study design which is going to be determined by the magnitude of the effect size you are expecting in addition to the number of variables you are trying to control. There isn't really a bog-standard, "this is the bare-minimum number of samples" that isn't arbitrary at some level. Of course, that's not how a lot of science is practiced, but even then it's possible to design a study ahead of time that's well powered without taking it into account.
It's been theorized for a long time that elephants are able to communicate via conducting sound through their feet, which seem to be specialized to pick up very low-frequency sound waves. It's thought that elephants will even choose different kinds of terrain to stand on to be able to listen or rumble better.
For those interested in the mechanics of this, there is some college-level biology stuff here with one image in particular that I was hoping would be a bit more detailed, but at least the rest of the document seems to explain it well.
There are also some YouTube videos about this phenomenon (one TEDx by Dr. O'Connell from 2013) that you can watch from this article.
Elephants are really cool and social creatures and we're constantly learning more about them and proving hypotheses!
npr
469 doesn't sound like a huge sample of calls, and how many of those have "name calls", and then 27% of the time guessed correctly is...how many instances?
Does it make sense for me to feel as skeptical as I am? What if the enthusiastic response is due to the call meaning something like "come here" and "oooh check this out if you're part of my herd " and "food for my babies" compared to non interesting things like "food but only if you're my baby". Or if the response is due to how much they miss the caller? Or even how muffled the recording is, making it a sort of, oh that's my mum but I can't hear what she's saying I better get closer.
They haven't isolated the "names" yet it sounds like. The next step would be if the researchers could reproduce just the name and use that to attract specific elephants? I don't know why I'm being so picky about this and not as excited about it as when I learned about bees wanting to play with balls
Disclaimer: I only have an entry level statistics and a biostatistics course for my background. All statistical data can be presented in such a way to support almost any conclusion. But I don't think that's happening here.
EDIT: Notcoffeetable and C-Cab's responses below are much more useful as to the discussion of the sample size aspect. My statement about 20 samples isn't really relevant here, and there's a lot more nuance to sample size than just saying there's a minimum number.
Generally, you need a sample size of at least 20 to have enough data for statistically significant predictions to be made, so 470 (ANOTHER EDIT: It's actually 101 unique elephants, 469 separate recorded calls) isn't too bad of a sample size for biology research. I just perused the paper itself very briefly, and they discuss some of your skepticisms in the Discussion/Conclusion section. Additionally, their control model consisted of ten thousand separate models that each contained nothing but random acoustic permutations, and none of those 10,000 models could identify receivers based on their calls more than 10% of the time. Their own model could identify 27.5% of receiver identities, and according to their statistical analysis in the paper, there is less than 0.01% chance of that happening due to a random coincidence (P<0.0001). I think it seems pretty likely that elephants are actually communicating with each other specifically in some way from this paper's data. Quoted below is also a relevant paragraph from the discussion section to your comment:
It's important to keep in mind that there is no standard number of samples you need that applies to every study. Estimated effect size is super important for determining the statistical power of any given study design, such that smaller effect sizes will need more samples to identify statistical differences. In my experience, a lot of physiological experiments can get away with fewer samples as a result of this, but behavior data needs a bit more as the variance is much higher.
You're right, thanks for pointing this out. I've edited my other comments to try and clarify and point towards you and Notcoffeetable's comments regarding sample size.
Going to need a citation on that. From what I know, good sample sizes range from 100 to 10% of the population size, but it ultimately depends on a lot of nuances of the study design.
Generally the number given in basic stats classes is about 40, some fields will accept 20 (in particular exercise science comes to mind where the interventions are invasive enough to make recruitment difficult.)
I am a mathematician not a statistician but the value 40 gets tossed around because I believe the standard techniques for computing p-values are well behaved. But I think (correct me if I am wrong) a statistician would not be happy with n=40 (much less 20) unless that was a decent portion of the population.
I work in a more statistical role, and generally I don't do much with 500 data points when it comes to ML modeling.
Edit: after a bit of reading, the number I'm seeing tossed around is n=30. And this is more or less a heuristic based on the Central Limit Theorem and its rate of convergence.
Sorry, I should have been more clear and mentioned that 20 is an absolute bare minimum and not an ideal number. ̶W̶i̶t̶h̶ ̶t̶h̶a̶t̶ ̶s̶a̶i̶d̶,̶ ̶a̶t̶ ̶4̶7̶0̶ ̶e̶l̶e̶p̶h̶a̶n̶t̶s̶ ̶w̶e̶ ̶a̶r̶e̶ ̶f̶a̶r̶ ̶b̶e̶y̶o̶n̶d̶ ̶t̶h̶a̶t̶ ̶l̶i̶n̶e̶.̶ EDIT: It's actually 101 unique elephants, 469 separate calls recorded. And when you consider how many resources it must take to track elephants and record what they do acoustically, 101 separate elephants is probably a realistic sample size. But also only having a sample size of 101 is probably a contributing factor to why their model is only identifying 27% of recipients. If they could've sampled more individuals, accuracy might be higher.
Regarding citation, I am trained as a medical lab scientist in the US, and was trained that 20 data points is the minimum needed when you're doing a validation on a new test that you're bringing into your lab, so that's where my number 20 came from. You need 20 data points from the new and old assays to compare (if you have a previous version of the assay.) Ideally, you get more than 20, but that's not how healthcare operates in the real world oftentimes 😆 Additionally, if your lab identifies the reference ranges for your own population (for example, what is the reference range for glucose for our patient population?), I was trained that you need at least 20 samples when identifying that. I understand this isn't really analogous to measuring unknowns about elephants, but I'm just explaining where that number came from.
Quick googling found this article, which supports having 20 samples for patient reference ranges towards the end of it, although additionally 95% of those samples have to line up with the published reference range for the analyte being measured: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10349316/
So yeah, 40 is probably a better minimum number lol
Worth noting: it's 101 elephants. 470 unique calls. But as a linguist, I don't think it's a bad sample size, especially to just explore the idea of whether these are distinct naming calls.
Whoa that's a big difference, thanks for catching that!
One important thing to keep in mind here is that sample size should be determined by the statistical power of your study design which is going to be determined by the magnitude of the effect size you are expecting in addition to the number of variables you are trying to control. There isn't really a bog-standard, "this is the bare-minimum number of samples" that isn't arbitrary at some level. Of course, that's not how a lot of science is practiced, but even then it's possible to design a study ahead of time that's well powered without taking it into account.