Marine science research test help

Hello stat experts.

I have been assigned with the data analysis on some marine science research, but I am by no means an expert. I was hoping to get some advice on what test I should use as well as how I should set up my data. I think I need to use a one-way ANOVA but am not sure. My research question is: Is there a significant difference in the amount of fish between different artificial reef types? The details of the data set are:

-5 different surveys, each split into two days (the first day is at location #1, the second-day location #2) over a period of months between 2004 and 2005.
-In each survey, there are 5 different reef types, with two repeated measures (amount of fish) for each reef type.

I would also like to test to see if there is a difference between the two locations as well. I am no data scientist so any feedback would be very helpful! Thank you.


Fortran must die
You can use either regression or ANOVA. Why did they split them over two days. I don't see a before or after aspect to this.
Sorry, I didn't explain that part very well. There are 5 surveys total. For example, Survey 1 is over two days (day one is at one location and day two is at the second location). That goes for all the other surveys. There's no really before and after. It's just to see which group/reef type has the most fish.


Fortran must die
The problem with that is you have to assume day does not matter as a separate factor. If it does you have to address it. It would have been a lot simpler to measure both locations at the same time.

The problem now is like this.
You measure one of the locations on day 1. You measure the other on day 2. Conditions are different on day 1 and 2. Now something other than the two locations have to be addressed.
I definitely agree, but unfortunately, I had no say in the experimental design. Your point is very clear, thanks. So now, as you said, we have 2 factors: Reef type and location/day or would it be three factors (Reef type, day, and location)? In that case, would I need to use a two way ANOVA? Thanks.


TS Contributor
So you have 10 surveys, and each reef type was surveyed twice, at different locations?
Since your dependent variable is a count variable, I do not understand why analysis of variance
here, except you assume n=10, but then for each reef type you'd have n=2 only.

Reef type and survey and day seem inevitably confounded, but that would be
the case with any analysis. I thought about just doing a one-sample Chi² with
reef type as categorical variable. But I may understand the study design incorrectly.

With kind regards



Fortran must die
That sample size is going to be a major issue with statistical power and probably generalizability. Ignoring the confound issue (I honestly don't understand why they designed it that way, it builds a confound into the data that is entirely unnecessary).

Rather than run regression (since it is count data I would guess Poisson regression would be the right approach although I know little of that) given the little amount of data, maybe you should just do descriptives and explain the issues with doing regression. If n is really 2 or even ten I am not sure the regression will even run (or ANOVA its the same thing).