# Compare two methods with location dependent values

I want to compare two counting methods (binoculair and nightvision). At multiple locations people counted animals simultaniously with both methods.

However, I don't know which statistics to perform because the locations are very different. Can anybody help me?

The data look like this

location method no. animals habitattype disturbance
A Bin 4 Grassland None
A Night 6 Grassland None
B Bin 20 Forest People
B Night 36 Forest People

When asking questions like that it is usefull to provide some example data. That could look like this:

expl <- data.frame(loc = rep(c("A", "B", "C", "D", "E"), each = 2),
method = rep(c("bin", "night"), 5),
count = c(2, 4, 40, 48, 14, 21, 33, 34, 18, 30))


which is

> expl
loc method count
1    A    bin     2
2    A  night     4
3    B    bin    40
4    B  night    48
5    C    bin    14
6    C  night    21
...


A very simple model to start with might look like this: Every location has a number of animals to see via binoculars and a constant number of animals more that can only be seen with nightview. Does that reflext the truth behind the biology? Probably not, but it is easy to compute via a very simple linear model and those are often a good place to start with:

mod1 <- lm(count ~ loc + method -1, data = expl)
summary(mod1)


In our example that leads to the following result:

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.370e-15  2.480e+00   0.000 1.000000
locB        4.100e+01  3.202e+00  12.806 0.000214 ***
locC        1.450e+01  3.202e+00   4.529 0.010585 *
locD        3.050e+01  3.202e+00   9.527 0.000678 ***
locE        2.100e+01  3.202e+00   6.559 0.002794 **
methodnight 6.000e+00  2.025e+00   2.963 0.041423 *


So we have a number of animals per location and an additional number of 6 animals that we expect more with method == "night". This number of 6.0 animals is significantly not zero with $$p$$ = 0.041.

Now we could improve on this very basic model with an interaction term and with poisson regression and prior assumptions and who knows what, but that will depend heavily on your statistics knowledge. As a first step, find a model that describes your idea of how the data came together that you can ideally express in terms of some form of linear regression. Linear regression is where you start and what you will continue to do amongst other things for the rest of your scientific career.