0
$\begingroup$

I am currently running some linear models and generalized least squares in R to detect latitudinal effects in the body size of a group of marine invertebrates.

The response variable is a measurement of the body size per each species included in the original dataset. The explanatory variables that I included in the models are the following ones: - Latitude (both absolute and squared values, to detect both linear and hump-shaped relationships). - Species records (a likely confounding variable that reports the number of records of species per latitudinal band of 1º). - Species richness (a likely confounding variable that reports the number of species per latitudinal band of 1º) - Sea surface temperature (an environmental variable that varies through the latitudinal gradient). - Net primary productivity (an environmental variable that varies through the latitudinal gradient). - Ocean (it's a categorical variable with six values: W Atl, E Atl, W Pac, E Pac, W Ind, E Ind). It represents the main coastline where the species is present.

I perform a first model, only including latitude (both absolute and squared values), the confounding variables (species richness, species records) and the ocean. Then, I perform a second model including latitude, confounding variables, ocean and environmental variables (sea surface temperature and net primary productivity).

I initially performed the models with the whole dataset (data from both the northern and the southern hemispheres), and I got statistically significant results. However, when repeating the models separating the original dataset per hemisphere (analyses of the northern and the southern hemisphere separately, to detect in the observed trends are present in both or only in a single hemisphere), I didn't get significant results. How is it possible?

Is it possible to obtain results for the whole dataset but nothing when separating per hemisphere?

$\endgroup$
0
$\begingroup$

Without seeing the data it's difficult to say, but surely the sample size is larger when you have the entire dataset, and you just don't have the power in a smaller sample? There are other ways you could handle lat/lon, e.g. Why does including latitude and longitude in a GAM account for spatial autocorrelation?

$\endgroup$
1
  • $\begingroup$ thank you for your response. I have about 1400 individual occurrence data, but most of them (ca. 85% of the total data) are from the northern hemisphere, so this one is well represented. However, ca. 100 of the occurrences belong to the southern hemisphere. That's why I expected to get similar results of the northern hemisphere compared to those of the total dataset. But maybe it's just what you said: the power is not enough for an smaller sample. I also attached spatial autocorrelation in the models. $\endgroup$ Mar 2 '20 at 13:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.