My partner and I are planning to count the ectoparasites found on a specific bat species, namely, Eonycteris spelaea, and find a correlation with the ambient temperature. We will be counting it twice in one year. We are planning to use Regression and Correlation, and were wondering if there would be any other more accurate approaches for our study.
Hi Angelica,
As you are using parasite counts/individual as your response variable, it might be appropriate to use a generalized linear model with a Poisson link. While this method is typically used for counts of events within equal time intervals, it is applicable if your measurement technique is the same for all individuals. You would have to include any measurements that might affect the number of parasites, perhaps age of bat, density of roosting, etc. as well as the temperature. The parasites themselves might be sensitive to temperature, so I think that your model will have to be comprehensive to provide useful results.
Hi Angelica,
As you are using parasite counts/individual as your response variable, it might be appropriate to use a generalized linear model with a Poisson link. While this method is typically used for counts of events within equal time intervals, it is applicable if your measurement technique is the same for all individuals. You would have to include any measurements that might affect the number of parasites, perhaps age of bat, density of roosting, etc. as well as the temperature. The parasites themselves might be sensitive to temperature, so I think that your model will have to be comprehensive to provide useful results.
Typically this sort of thing is done using a "growing degree day" (GDD) model. Insect development is related to the accumulated heat units, so each 24 hour period above a critical threshold developmental temperature will allow the insect to perform a specific number of metabolic tasks. When enough of these are finished, the insect can develop to the next stage. That said, regression analysis should work great, unless there is more to this project than indicated in the question.
The GDD model needs continuous monitoring of temperature. One measurement every few minutes works great, though one could make less frequent measures work. I am not sure what the minimum is, though I would think you need multiple measurements each day. One measurement every six months would not work for this.
Since you have a count data, I suggest the Poisson Regression Model.
You should consider sex of the individual, body weight, if the bat is pregnant or not, and nursing. These simple observations could greatly improve your data. Of course there are many other factors that could be measured if sufficient resources were available. Jim's comment about being consistent is very important.
It may be that the bats can behaviorally regulate temperature (roost closer together as it gets colder) and this would diminish the effect of ambient temperatures on ectoparasite biology. So ambient temperature, cage temperature, and roost temperature could be important. I am not sure how much individual temperature varies. Some people run hotter than others, and I would assume that this applies to the bats. Also some people are preferred such that 9 of 10 mosquitoes prefer Tom over Cynthia. I have no idea how much the equivalent differences in bats will be important to this ectoparasite.
i also think Poisson regression model will work,all the best in your research
What do you consider the "ambient" temperature?
Outside the cave?
Inside the cave?
Where the bats roost?
Where the bats hunt?
Only on the day you sample?
I agree with Tim. You should be looking at more than one factor in your analysis. Poisson Regression is probably going to be the proper method to use. I could see Time Series Analysis coming into this project too.
I'd also add colony of bat and spatial data to the model.
Good luck!
Hi Angelica
As many mention above, Poisson regression would generally be the preferred approach. If there are more zero counts than expected for a Poisson model, try negative binomial regression, or alternatively zero-inflated Poisson or zero-inflated negative binomial.
You don't give many specifics about your design. Andrew Ekstrom points to several things you might also want to consider, which could lead to correlated ectoparasite counts between individuals and thus would cause non-independence. For example, are the bats sampled at multiple roosts? If you are sampling bats during the breeding season, perhaps you need to account for family groups (etc.)? If any of those sorts of situations apply, consider using Poisson or negative binomial Generalized Linear Mixed Effects Models (GLMM), where you could account for correlated effects by treating e.g. roost or cave as a random effect to account for the non-independence of samples collected at the same location (or family group). An advantage to that approach as well is that you can estimate variance components.
Cheers
-Steve
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis