Which statistical test to run?

#1
Hi everyone,

I conducted a study on 20 captive elephants looking at time activity budgets. Behavioural sampling was conducted through the conspicuous behaviour recording. Since my focus was on behaviour I have the duration for each of the 8 behaviours I was looking at, for all 20 individuals. Sampling was conducted between 9:00 am- 15:00 pm- with the bulk of the observations conducted between 11:00 am- 13:00 pm. My challenge is that (i) my sample size is quite small (n=1985 minutes) (ii) the my time spent on observing each of the 8 behaviours is not equal (because it was focused on behaviour).

I hypothesised that certain behaviours will occur more frequently during certain hours as compared to other behaviors. Since I have a very basic understanding of stats I would be very appreciative of any advice on what tests would best suit my study.
 

Karabiner

TS Contributor
#2
my sample size is quite small (n=1985 minutes)
I'd suppose that your sample size is n=20 elephants.
I hypothesised that certain behaviours will occur more frequently during certain hours as compared to other behaviors
Could you be more precise here? For example, what do you mean by "certain hours"? What are these "certain behaviours"?

What is your dependent variable here, how does it actually look like?

With kind regards

Karabiner
 
#3
Thank you for getting back. My dependent variable would be the behaviours (viz feeding, play, bathing, dust-bath....). I assumed that feeding would take place during the mornings between 10:00-12:00- and the other comfort behaviours (dust-bath, bathing, sleeping etc) would have a higher occurrence during the afternoon (post 12:00/ the hotter parts of the day). I recorded the duration of the behaviours and have created a proportion of time spent on each behaviours (attaching figure) by the population. I am also attaching the graph showing occurrence of behaviour during certain parts of the day.

The concern is that the large part of my scans occurred during the 10:00-13:00 (due to the animals being in a visible location), with very few scans done from 9:00-10:00 and 13:00 to 14:00 (where they were not present or they were out of sight). I feel that there would be a bias in my data since scans were not equally conducted across time periods. Looking forward to your advice.
 

Attachments

Karabiner

TS Contributor
#4
Since you analyse proportions instead of actual time
spent, there is perhaps no bias, as far as I can see.

But if "feeding" is associated with "the animal can be seen"
(or maybe with "...cannot be seen") during hot hours,
then you would have a bias.

Seemingly, you have a dependent variable "proportion
of time spent", measured with respect to feeding and
to non-feeding behaviour. Your hypothesis is still not
completely clear to me. If you want to analyse whether
feeding has a higher proportion in the morning
than in the afternoon, then you can perform a
dependent-samples test (e.g. Wilcoxon signed rank test).
If you want to test whether feeding has a higher
proportion than non-feeding in the morning, then
you could perform a dependent samples test (such
as the Wilcoxon).

If you want to know whether feeding mainly takes place
in the morning and not during the afternoon (something
like "90% of feeding between 9:00 a.m. and 2:00 p.m.
is done during the morning"), then perhaps you would
have to construct a new variable .


With kind regards

Karabiner
 
#5
Thank you for your response. Since this was the first study that was conducted on behaviour patterns on this population of elephants, to be honest I really didnt have a hypothesis- it was more of an observatory study. I just wanted to see whether the captive elephants engage in behaviors that wild elephants do and during which parts of the day. The understanding was that the behaviors that we see would be similar to wild elephants. Since this was my first study that I conducted the methodology is not very good. However, with the data that I have I was wondering whether it would be possible for me to run any inferential stats on it or should I just keep my results as descriptive?
 

Karabiner

TS Contributor
#6
For those hypotheses I could imagine, I suggested an appropriate test.
But maybe you should do some thorough descriptive statistics first,
which gives you more insight into the data.

With kind regards

Karabiner