UKH

If one wanted to test for any significance regarding which compass direction swift nests are orientated, which of the more straight forward stats tests would one use?

Essentially, it's more about 'being seen to do something intelligent with the data' given the depth in which I'm able to look into the nest properties. 

Many thanks.

I guess the simplest would be a 4x2 Fischer's Exact table with the NSEW values compared to the expected (even) distribution.

Are you using degrees or grouping by direction (eg 12 nests fall in the "north" category, vs nests with orientation 320deg, 10 deg, 13 deg.....etc)? this may affect the type of test you can use.

I'm guessing you have categorised?

Have you done something like a wagon wheel or pie-chart for starters?

Just to say your dissertation sounds really interesting, be great to hear of any findings etc etc. Also, I’m aware of a campaign about installing ‘swift bricks’ in new building houses. All good stuff, swifts are amazing birds. 

Timmd....long time no see! Hope you are doing well...best wishes!

If you haven't already read it, you might be interested in this:

https://www.theguardian.com/books/2023/jun/16/one-midsummers-day-by-mark-co...

It's a fascinating take on the world that swifts live in, and hence makes one think about one's own place in the world and the interactions we have with the natural world. 

Thank you, and to you. I more or less figured life needed my focus rather than UKC, would think about logging on, and then think about something else.

Could a chi square test work, with an equal weighting for NSE or W? 

How many nests have you measured?  Tens, hundreds or thousands?  

It is 52, could a chi square test fit with a null hypothesis of equal number of nests (13) facing in each direction?

A chi-squared test will only tell you if there is a bias (ie they not are distributed to an assumed pattern, such as random or equal compass points). To say how much they prefer facing (say) east and what additional factors (variables) may explain direction, you may consider multinomial regression modelling (especially if your data is literally N S E or W) or linear modelling with a circular outcome variable (if you have full compass degree measured).

See https://stats.stackexchange.com/questions/148380/use-of-circular-predictors... for starters.

Without going beyond A Level maths standard, you could do 4 different binomial hypothesis tests, first test if they are more likely to point north, then repeat for each direction

1) assume X~B(52,p)

Ho:p=0.25

H1:p > 0.25

This might get shot down by more educated minds on here, as it reeks a bit of fishing around for a hypothesis which works, but my stats knowledge doesn’t go much above this level so it’s all I’ve got!

It was actually 62 nests, but the fewest nests facing into the prevailing winds possibly correlating with other research suggesting lighter winds favour brood survival is quite interesting. 

There are various ways you can compare your null hypothesis to your measured data.  There are multiple ways to do it with chi squared (or reduced chi squared).  What's appropriate really depends on the data.

My next question is why you say "each direction" as if N, S, E and W are in any way important - we choose to break our compass in to four primary directions (I’m assuming this is about wind not magnetics which do impose some absolutes…).  Imagine if the nests all pointed NE - you'd get a very different result if your null hypothesis was equally facing N, S, E & W vs NE, SE, SW & NW.  If you force some measurement buckets on to the problem, you limit the ability to interrogate the data.  Four buckets feels too small for this.  You can probably get away with 8-9 buckets with 52 samples.

Another question is to what precision do you measure the angle?  1°?  90°?

My first port of call would be to look at the angular difference between every nest and every other nest.  If there is a uniform distribution of nest angles, this will be uniform.  If there is a preferential direction it will cluster.  When comparing angles, you can go the long or the short way round the circle.  Always go the short way, e.g. difference_between(350°, 10°) can be either 20° or 340°.   Measure the "shortest way round" angle between every nest and every other nest, then look at the distribution of these values.   You could plot a histogram or you could plot them in ranked order by magnitude.  Get a feel for if there's anything actually happening before you design a statistical test for it.

There's also The Screaming Sky by Charles A Foster. Fascinating and also slightly barking mad.

The author also wrote Being a Beast wherein he tried to live like animals e.g. by excavating a sett with a JCB and eating worms.

That's a very intelligent and considered response.

Currently, pending a mark I'm yet to get back, my grade average is in the 2.1 band, so I'm following a marginal gains approach in trying to get any mark I can do where I can do to keep it there.

Other modules play to my strengths more, along the lines of grasping concepts and absorbing  related information, and writing about or doing screen cast presentations about them.

So it's more along the lines of 'marginal gains' that I'm after a way of crunching the numbers in an intelligent seeming way, though some papers I've seen in journals have only talked about the broad N S E and W compass aspects too. 

(In the worst case, if I get a 2.2 and that finds me a job I can be content in, I may do another online degree in things I've broadly already done towards a 2.1, if I ever reach a point where I feel it's holding me back.)

It turns out to be 62 nests...

If you send me the data I’ll give it to my Y13 kids as a homework.

For the avoidance of doubt im not joking, they will probably do something similar to what I suggested above but it might be enough to scrape you an extra marK and they get to deal with a real data set.

Thanks very much, you'll have an email by the morning. 

African swifts* or European?

*Ref. a Monty Python joke

1) not a statustician.

2) presumably there is no pilot to create a hypothesis.. did you have a specific question before you started collecting data?

3) remember, you need a result of practical as well as statistical significance (obv must be statistically significant but you have an end goal for this.

You need to define a question to answer...

You only have 62 points and the overall direction is going to be approximate. BEFORE the next step  you need to decide how divide the data - quarters or sixths probably.

Then do a bit of exploratory data analysis (this is acceptable)....graph the results x is bearing in degrees y is frequency.

Given that '0' is arbitrary you can then decide where to start the cutting so you don't chop peaks up and dilute the result. (Shifting the zero by 45deg could abolish a 'peak  / trough')

You can have a null hypothesis that your test group is no different... chi squared etc. I know this tips into data dredging but it is a simple solution, and if the methodology is clear that should be OK.

Thanks, yes, I'm going to talk a lot about the limitations of my study and methods, on a good night's sleep, it's looking like given the limitations, that it could be just as scientific to say there's nearly equal proportions of nests facing in all directions except for facing East, and then explore any potential link between the prevailing winds, and swift fledgling survival rates appearing to correlate with gentler winds abutting the nest sites.

To mangle Rutherford...

If you need statistics to get a result you should have designed a better experiment....

(And you may well have a result in the data)

Yes indeed, it's West that the nests are notably not facing as it happens, rather than East. I didn't expect to find that.