I’m new to the publishing world and I’m hoping to get some advice on the most appropriate model to use to statistically analysing my data please. To summarise the project, I deployed four AudioMoths (which are audio recording devices) at four different sites in the middle of the Davies’ tree frog’s breeding season (one AudioMoth per site). There is virtually no published research on the species and the frog’s daily calling activity has not been described. The aim of the paper is to describe the daily calling activity of the species (i.e., what time of day does the frog call at) and I am hypothesising that the frog will peak in calling activity in the hours just after sunset (i.e., between 7-9 pm).
My AudioMoths recorded 5-mins at the start of every half an hour in a day for a 10 day period. I used an automated sound recognition software called Kaleidoscope Pro to extract the frog’s call from all the audio data. I then calculated how many calls there were in each hour of the day (so there’s two lots of 5 min recordings per hour) and then took the average number of calls per hour from all 10 days for each detector (so each hour has a sample size of 10, 1 hour for each of the 10 days). I then combined the data from all 4 AudioMoths to get an overall average number of calls for each hour of each day (so each hour now has a sample size of 40). I then turned the average number of calls per hour to a percentage of calls per hour and created the attached graph with standard error bars.
What I’m hoping to get help with is how to appropriately test if there is a difference between the number of calls in each of the 24 hours per day? Would I use an ANOVA initially to see if there is a difference in the number of calls between any of the hours and then follow it up with a post-hoc test to find out where the differences are? Or would it be best to compare the percentage of calls per hour and use a chi-squared test? Or is there another better option? What problems arise when I’m comparing 24 different groups (because there are 24 hours in a day)? I’m mainly wanting to know how best to show statistically that there is a peak in calling between 7-9pm (if there is in fact a statistical peak)? If there is a statistically significant peak in calling between 7-9pm, then future field surveys can survey in this time period to maximise the probability of detecting the species.
I’d love to hear your thoughts on this if you have the time please! I’d also really appreciate to hear your explanation as to why you suggest the approach you do please?
Factorial experiment should be analyzed by SAS, SPSS statistical software
I can analyze your data at minimal cost and help you in less than 24 hours
Hi there Lachlan,
I don't think you want to do ANOVA or similar linear models since you're using time-dependent data (the number of calls at one hour depends on the number of calls the previous hour). I think your approach of averaging throughout the days is good, but maybe there's no need to average within the hour. Why not have 48 points instead of 24?
Regarding the analysis, you need to do something called time-series analysis, which can get quite complicated (https://stats.stackexchange.com/questions/30640/a-list-of-common-time-series-tests). Instead of going deep into time-series statistics, I propose you perform pair-wise t-tests for each step (0 vs. 0.5, 0.5 vs 1, etc.). There will be significant differences if the frogs start/stop singing. Looking at your graph, that'll be significant at four different times (18-19; 20-21; 5-6 and 7-8). If the difference (t1 - t0) is positive, the frogs are singing more, if it's negative, they're singing less. Using the significances and differences you'll be able to describe their singing behaviour.
Just a note, you might want to consider plotting the counts in a circle plot (polar coordinates), since the end and beggining of the plot are the same hour (24/0). Here you can see some examples in R. http://rstudio-pubs-static.s3.amazonaws.com/3369_998f8b2d788e4a0384ae565c4280aa47.html
Good luck!
- Badges
- Science topic