Guidance on appropriate analyses

[#1]

I am trying to relearn JMP and the appropriate stats to use. I would really appreciate help trying to determine which analyses are most appropriate given the dataset. I have a dataset of whale sightings collected from aerial and vessel surveys with 3 different environmental covariates [distance from shore (km), depth (m), and Sea Surface Temperature (SST degrees C)].

 I have attached the summary stats for the (1) covariates only and, (2) for the covariates X Species (‘common name short’).

Generally, the data are not normally distributed and heteroscedastic for at least distance from shore, and depth. Also variances are not homogenous therefore I expect nonparametric tests are the way to go.

The questions I am interested in are as follows:

(1) Is there a significant difference between species for three different variables?

--distance from shore

-- depth

--SST? 

(1a) Is there a seasonal effect of distribution based on the three variables above. (I.e. between winter, spring, summer, fall)?

(2) Is there a significant difference in group size between seasons?

(2b) Is there an effect of species on group size by season?  

I would sincerely appreciate guidance as to what is the most appropriate analyses are given the data and how to properly report the results from the chosen tests.  THANK YOU!!

[#2]

Welcome to the community. I have some questions and then some guesses. How was the data acquired?  Can you attach the JMP table?  Assuming the response is count data, how many sightings are there?  How confident are you in the accuracy of the data? Is the seasonal data and time series in the data set?  I always start by looking at the data.  Try Graph Builder and you can drop the variables into the drop zones to arrange the data many ways. Then maybe logistic regression (or standard least squares if there is enough count data).  Also check for multicollinearity (VIF's or Multivariate Methods>Multivariate) and outlier data points.

[#3]

Take a look at this example. Its similar to your problem, might get you going in the right direction.

https://www.jmp.com/support/help/en/15.2/index.shtml#page/jmp/simple-split-plot-or-repeated-measures...

[#4]

I thank each of you so much for the feedback! I can't tell you how grateful I am for this guidance as I'm reteaching my self and feel quite naiive and buried in all the rabbit holes I've dug into, at this point.

I hadn't considered repeated measures so I am looking more into the example with the link that was sent. I expected non parametric tests would be the way to go so I will look into these other suggestions. Thank you!

I am very confident in the data. It was all systematically collected, managed and QAQCed based on standard protocols across the almost 20 years of the study. The only data that wasn't directly collected is the environmental data. It was annotated from Movebank a reputable and respected site for environmental data and animal tracking data (https://www.movebank.org/cms/movebank-content/envdata-products). Not all sightings have associated environmental data.

I

The questions I have in regards to the most appropriate analyses. 

1.  determine if there is a difference in group size between seasons. 
2.  determine if there are differences between species relative to depth, distance to shore, and SST.

Regarding the point on outliers, I also wondered about this with regards to the environmental data. There are a few points that are exceptionally large for depth, but there are several records in the 3 thousand meter depth for 3 different species that are known to spend time offshore near deep water canyons. So I am cautious in removing these 'outliers' as they may very well reflect ecologically interesting and valid observations. 

 Thanks again! 

[#5]

Sorry, I'm tied up today.  Why don't you attach JMP files?

My question about the validity of the data includes how do you know the same whales are counted more than once?  My suggestion of looking for outlier data points is not so you can remove them.  They are often the most informative part of a data set, but you do need to realize they can distort quantitative analysis.

[#6]

I'll let others respond regarding non-parametric tests. But I would at least try log transformations (which should make the distribution more well-behaved) and then ANOVA (take the negative of the depth before taking logs, of course). Even if the variances are unequal, I have rarely (very rarely) seem that it has a marked effect on the results to do a more sophisticated test. It also seems that you should want to investigate combinations of the covariates, so fitting a model rather than Fit Y by X would be my next suggestion.