Jump to content


Correctly Determining Sexual Dimorphism In A Shiner


16 replies to this topic

#1 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 16 June 2009 - 05:16 PM

I have a question that's really a form of extended peer review, I guess. I have a submitted manuscript on the timing of reproduction of telescope shiners in Hurricane Creek, AL. The reviewers' comments came back last week, for the most part with good ideas of what to change and how to fix the boneheaded things I left in myself (the way of the world!). Anyway, in the manuscript we report on observed size of the adult telescopes that we collected by seining in roughly equal monthly numbers from February through September, roughly 35 a month. By pooling all of the males together, and all of the females together, the mean standard length of the 162 males was 45.4 mm (S.E. of 0.3) and the standard length of the females was 50.5 mm (S.E. of 0.5). An F-test found no significant difference in variance between the males and females, and a t-test comparing males to females showed a significant difference with P<0.01. So, we reported this in the paper as evidence for sexual dimorphism, with larger females than males.

The reviewer's comment was, "Unless you know that you are examining fish of the same age class, the approach to sexual dimorphism is unconvincing. I think it detracts from the other portion of the manuscript and should be removed."

From looking at length frequencies of individuals from March, June and September, our interpretation is that most breeding individuals are two years old, with a few three year olds. These fish just don't live that long.

So my question to everyone who's read this far is, does our comparison above with the t-test convince you that we've truly demonstrated sexual dimorphism in standard length? I'd rather not drop this from the paper purely as a natural history observation. I did the same thing in a recent paper about burrhead and silverstripe shiners and neither the guest editor nor the two reviewers said boo about it.

Thanks for any feedback.

#2 Guest_farmertodd_*

Guest_farmertodd_*
  • Guests

Posted 16 June 2009 - 05:48 PM

I don't understand Bruce. You're reporting the mean SL (45.4, 50.5) for all the fish you collected over all the months? That's all samples, all months, every SL you measured? If that's so, I'd knick you for not controlling for time, the samples aren't independent, thus no t-test (and I'm pretty lenient about stats in natural history :) ). You'd need to do a repeated measures ANOVA, and that's probably more trouble than what you want to go through.

You could test each month against itself with t-tests, but this is t-testery run amok :)

Maybe I'm not understanding this right. And I do understand you wanting to put it in as a FWIW because it's possibly biologically significant, but I don't know that I would call that statistically significant.

What have others done with dimorphism?

Help me understand this better. I think it's an interesting problem to solve.

Todd

#3 Guest_centrarchid_*

Guest_centrarchid_*
  • Guests

Posted 16 June 2009 - 06:18 PM

Bruce, I have an opinion in the context of my aquaculture interest. Since growth of the telescope shiner is indeterminant, a comparison of standard length as a function of gender across such a large time scale provides very limited information I could use in making management descisions. What might be more informative is the realitionship between standard length, month and year class that could be presented as 3-D scatter plot. Regression analysis could be used to predict length at age that might ultimately be related to efforts for sorting brood animals or modeling a predator size/ prey size relationship as a function of age and sex of the shiner.

#4 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 16 June 2009 - 06:23 PM

I was afraid you said you'd kick me, not knick me... You're right, I'm trying to do it the easy way without handling or sacrificing several hundred from, say, June. If you look at each month's SL data they vary widely, which makes sense with a sample size of ~35. Is it random sampling from the same population of a smallish creek over 8 months, even with mortality happening from various causes? Or are these various causes enough to make each month a truly independent sampling population which is certainly the safer, conservative, dare I say accurate method? I've seen it done the way I did in various publications, but that doesn't mean it's truly the best way to do it.

#5 Guest_farmertodd_*

Guest_farmertodd_*
  • Guests

Posted 16 June 2009 - 06:59 PM

I would only ever knick you in peer review. There have been some people I wanted to kick, but that's mainly only because they work at the single ecological scale of "a plastic bucket" and that somehow translates out across the landscape how? ;)

Each statistical population is not independent, even if you do it month by month because it's the same locality (you'd be in much better shape if you hit 30 similarly sized catchments, but who the heck has time for that, so more than 5 would be great). You can make a comparison within a month, but you can not compare across months without controlling for time. If you control for time, you will get some independence back.

So the model might be something like: length = sex + month + (sex*month)

This way you're controlling for both sex, month and the mixed effects of month and sex.

You just took the first 35 you got and stopped, right?

I'm going to guess you did some histology on the gonads? Perhaps select the month where they were most developed and compare that month. Then it's just a statement "Female SL was significantly larger than male SL at the time of prime gonad development" and leave it there, perhaps as a question to delve into later.

That is unless you want to control for time. Then you could start to talk about length as a dimorphic characteristic for telescope shiner.

Do you have 30 of each sex for each month? If you do, you could talk a little more about dimorphism.

Otherwise, you're way ahead of what your data may suggest. I do this myself all the time :)

Hopefully, you find this helpful. I just went through all this crap with my masters. A LOT of people do this incorrectly, and it's taken me so long to submit for publication because I never believed the answers I was getting.

Todd

#6 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 16 June 2009 - 07:19 PM

I'm going to guess you did some histology on the gonads? Perhaps select the month where they were most developed and compare that month. Then it's just a statement "Female SL was significantly larger than male SL at the time of prime gonad development" and leave it there, perhaps as a question to delve into later.

That is unless you want to control for time. Then you could start to talk about length as a dimorphic characteristic for telescope shiner.

The name of the game is to be ahead of your data, then to rein yourself in. I have a total of 35 individuals, period, from each month as a condition of our collecting permit from Alabama Forever Wild. But June was the peak reproductive condition month gonadally, and we have a nearly even split M/F, so I might try that as a passing observation.

There's a whole literature mentioning sexual dimorphism in cyrprinids (see Mark Pyron's reviews, especially) and the measures are done in various fashion. I'll have to revisit that question. It touches on a lot of reproductive biology, but I don't want to have to sacrifice a large number of individuals at any one time for an answer everyone will agree with. We'll see...

#7 Guest_farmertodd_*

Guest_farmertodd_*
  • Guests

Posted 16 June 2009 - 07:48 PM

Yeap, I totally understand both limiting situations. Hind sight being 20/20, you might have thought to have a sample as an analogue from Tennessee where they're much more abundant and distributed. Or you might even just suggest that in your discussion. Or hang on to this study for one more season, reapply with the data you have to make the case for one more June sample (30 males, 30 females), perhaps non-lethal sampling, and get the corresponding data from TN.

There's a lot of ways you could go with it. It really depends on how important you think this tidbit is and what amount of effort you think you should put into it. Obviously, not all species will be abundant everywhere, so for life history work, there may have to be some angles that are overlooked in a statistical sense because of sample size.

You could also just say which months mean female SL was larger than mean male SL for all months and just steer clear of the whole dimorphism aspect (a leading conclusion, instead of making the conclusion for the reader) and hang this one up on the rack to dry.

Todd

#8 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 16 June 2009 - 08:01 PM

Telescopes are one of those silvery shiners where it's difficult to tell males from females non-invasively. Since we've already executed a bunch and looked closely at their gonads, it should (in principle) be an easy spinoff to address size dimorphism. Non-invasive examination and measurement would almost be easy with, say, scarlet or tennessee shiners in the same system where males are obvious. We've been working this past year with telescopes and scarlets in Estill Fork, the next stream to the west of Hurricane Creek, which has much easier access (paved roads!) and no requirement for special collecting permits. We'll have to regroup if we want to pursue the sexual size dimorphism question with telescopes.

#9 Guest_ashtonmj_*

Guest_ashtonmj_*
  • Guests

Posted 16 June 2009 - 08:32 PM

In short, I'm going to have to agree with Todd on the statistical violations. Maybe if you had something like three collections pooled from consecutive months and called it age/length relationship for summer you can get by but not something collected over a season. Some other things you could be looking at are fast growth in males, not putting as much energy into reprodcution as females, and then a cesation at maturity. So you could have 2 year old females that are smaller than 2+ year old males, thus not same age class. I think sample size will be too low for a length at age over time or with time as a covariate, though it might be worth a look. I think Todd's model is something along the same lines where you have the ineraction effect taken into account. Don't have it handy since I'm in the field right now and just wrapped up hour 16 for the 8th straight day, but with larger sample sizes you can do a kernel analysis to really get something much more statiscially powerful than a length frequency analysis, since we all know that those are never normal, univariate, etc. and almost always some polymodal mess. Bruce, just because there isn't an F, T or p to report doesn't take away from the fact you have valuable data to report. There are just some things in ecology that almost shouldn't be crammed into the mathematical realm of normal statistics. I think if you got more telescopes, from other populations, and developed a model at a larger sample size that accounts for them across their population you'd have a great paper along the lines of 'non lethal determination of sex in cyprinids'....

#10 Guest_farmertodd_*

Guest_farmertodd_*
  • Guests

Posted 16 June 2009 - 08:49 PM

Bruce, just because there isn't an F, T or p to report doesn't take away from the fact you have valuable data to report. There are just some things in ecology that almost shouldn't be crammed into the mathematical realm of normal statistics.


And that is dead on. Now convincing reviewers and editors of that is a whole other matter (I'd hate to try and get this through my department :rolleyes: ), but it sounded like your reviewer wasn't all that hung up on it. They just didn't want you to say that was a dimorphic characteristic based on the data you have (which I think you've come to the same conclusion at this point too).

Todd

#11 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 16 June 2009 - 09:06 PM

Well, yeah, I'm convinced that they are sexually dimorphic for size but it's not an ironclad conclusion. We're seeing the same relationship from fish collected in Estill Fork over the last year. It's one of those things that I'm not prepared to pursue in a big way at the moment. We're trying to sort out some other offbeat ecology, such as the telescopes from Hurricane Creek have a moderate load of Dactylogyrus gill flukes (n~40) but the ones we've examined from Estill Fork have none (n~20). But that's a whole new thread waiting to happen.

#12 Guest_ashtonmj_*

Guest_ashtonmj_*
  • Guests

Posted 16 June 2009 - 09:11 PM

Speaking from experience from a certain journal with the word 'Naturalist' in its title, where a reviewer essentially gave me a similar 'nice study, but not worthy enough for publication here (because there were no fancy statistics), so no thanks'....

What Todd mentions has to be one of my biggest problems when writing initial drafts. Say what the data says and only what the data says. I think you know that if you had larger sample sizes or multiple populations that your pattern would hold, but that data just isn't there to support it yet. Enough authors speculate or stretch things beyond their appropriate limits. If they are 2 mm larger, they are larger. That was just a fact of nature in the population you measured. It doesn't need a p value if you want to report on the population. It would be a great teaser, naturalist note as you mention, to a larger multi population, controlling multiple extraneous factors, estimation/model of lengnth at age to determine sex for a non-dimormphic species article....or it sounds like a nice 'based on previous work (Stallsmith unpublished data) it may be possible to determine sex ratios using length at age relationships' for a funding proposal.

#13 Guest_farmertodd_*

Guest_farmertodd_*
  • Guests

Posted 16 June 2009 - 09:46 PM

What Todd mentions has to be one of my biggest problems when writing initial drafts. Say what the data says and only what the data says.


I think this is suffered by most naturalist types entering modern North American science. You'd think coming in well informed and experienced in nature would help you out studying nature... And perhaps it does in the long run, but not until you learn to quiet all the noise, and get exactly to the question. Man, I have really struggled with this, and after I'm out and it's me vs the reviewers without the luxury of others trained in the process vested in my success, I may yet have to learn to remind myself of my tendencies.

Really honks you off too when you're like "I know I'm fricken' right, you just don't understand it the way I'm communicating it."

#14 Guest_Doug_Dame_*

Guest_Doug_Dame_*
  • Guests

Posted 16 June 2009 - 09:59 PM

The reviewer's comment was, "Unless you know that you are examining fish of the same age class, the approach to sexual dimorphism is unconvincing. I think it detracts from the other portion of the manuscript and should be removed."

From looking at length frequencies of individuals from March, June and September, our interpretation is that most breeding individuals are two years old, with a few three year olds. These fish just don't live that long.

So my question to everyone who's read this far is, does our comparison above with the t-test convince you that we've truly demonstrated sexual dimorphism in standard length? I'd rather not drop this from the paper purely as a natural history observation. I did the same thing in a recent paper about burrhead and silverstripe shiners and neither the guest editor nor the two reviewers said boo about it.

Disclaimer: not a card-carrying statistician.

I interpret the comment to mean that the stats may be skewed because of possible presence of a few "wrong year-class" specimens in the samples.

One possible solution for that would be to add a data chart to the paper to visually address that point, so the data would (hopefully) show that age intermingling within sample was not an issue. I'm imaging paired box-and-whisker plots aligned along a time axis. (I'd align the time axis vertically and the fish length axis horizontally, I think that would make it easier for the readers' brain to focus in on the variablity of length and any outliers that might be present.)

OTOH, you said there were

a few three year olds.

Is that assessment based on visual clues, such as size ? If so, those individuals could certainly move the averages differently from sample to sample or by sex, given that the mix of 2nd-to-3rd fish would likely be randomly different from sample to sample. So it could be that a visual plot would show that the 3rd year fish were different. So ...

And/or
- exclude the 3rd year fish from the reported SL stats. That gives a more pure indication of the range of sizes of the normal breeding cohorts, which is what I think you are primarily trying to convey to the readers .... mixing in 3rd year fish contaminates those numbers with uncontrolled effects, namely the relative survival rates of the fish from Yr 2 to Yr 3, and possibly changing habitat choices also.
- report medians and the interquartile ranges, they'll be less susceptible to being skewed by a few 3rd year specimens. And thus also less susceptible to carping from careful and precise peer reviewers. (I use the term carping in jest, the issue raised is legit.)

HTH

d.d.

#15 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 17 June 2009 - 07:18 PM

Thanks all for the comments and suggestions (especially Doug's, I had to read it twice to grok the fullness). For simplicity what I think I'm going to do is to merely report at some point in the Results that collected males were in a given size range, and females were in a given size range, with no effort to ascribe statistical power in the report. The funny thing is that I think the cited reviewer's problem with my original report is maybe the least solid of several reasons not to make my original claim.

We've observed some sexual size dimorphism patterns in various species. Alpha male scarlet shiners are much bigger than sub-alpha males or reproductive females, for instance, which makes sense with their pattern of tournament spawning. And for a species like telescope shiners, with no pronounced coloration or behavioral differences (at least to human observation), it may well be advantageous for females to be bigger to produce more eggs. That's the basis of my interest, and that's the challenge to more thoroughly document any patterns of size differences.

#16 Guest_Gene2308_*

Guest_Gene2308_*
  • Guests

Posted 16 July 2009 - 06:12 AM

Disclaimer: not a card-carrying statistician.


Ditto. I was just going to comment that running a repeated measures test (like a Friedman) really sucks for getting significance also. I have seen data that appear to have an obvious pattern miss a p-value and get buried in someone's desk somewhere.

I personally despise what statistics has done to much of science and resent the approach of "first I am going to decide what stats I can run, and NOW I will think of something interesting to study".

Your data is very interesting! Is it common to have dimorphism in shiners (literature or personal observation) ?

Edited by Gene2308, 16 July 2009 - 06:12 AM.


#17 Guest_fundulus_*

Guest_fundulus_*
  • Guests

Posted 16 July 2009 - 06:48 AM

Yes, a large number of shiner species show some form of obvious (to humans) sexual dimorphism. Some, like the Hydrophlox subgenus of Notropis or Lythrurus species, show pronounced coloration of breeding males. Others show size differentiation with larger males or females for reasons having to do with reproductive success -- larger males can defend territories from conspecific males in some species (with "tournament" reproductive systems) or larger females capable of producing more eggs, often in species with some form of group spawning. North American shiners in general are famous for displaying a wide range of reproductive strategies and behaviors, with phenotype usually reflecting these strategies.

I should also say that my manuscript on telescope shiners has been accepted by American Midland Naturalist, I'll post it when it's in print.



Reply to this topic



  


0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users