After searching here (asking tons of questions), I have now managed to do some initial fitting of my distribution using the
fitdistrplus package in R. I need some advise if I am doing this correctly or not or does the analysis make sense.
I have date that records percentage of planting every week (cannot exceed 100% and cannot be less than 0%) in for three locations. Here's the data:
loc.x1 <- c(0.8,8.3,41.5,35.4,11.2,2.8) loc.x2 <- c(6,21,36,35,2) loc.x3 <- c(5.1,21.2,39.7,14.7,10.3,7.1,1.9)
Each vector start from week 1, week2....weekn.
I want to fit a family. To do this I did this:
library(fitdistrplus) plotdist(loc.x1, histo = TRUE, demp = TRUE,pch = 19)
plotdist(loc.x2, histo = TRUE, demp = TRUE,pch = 19)
plotdist(loc.x3, histo = TRUE, demp = TRUE,pch = 19)
To get an idea of which family distribution to fit, I did this:
My questions are:
(1) Looking at the Cullen-Frey graph,
loc.x2 data cannot be fitted by the families shown in the graph (the blue observation is not even present in the graph). Does it mean that none of these distributions can be used to fit my data?
loc.x3, what do the plot suggest? Can I use a beta family as a good approximation of underlying distribution?
The reason the vector length are different is because in each location, number of weeks to plant is different. For e.g. in loc 1, it takes 6 weeks to plant (hence 6 values). Each value gives the % planted in that week. Similarly, loc.x2 planted in 5 weeks (hence 5 values).
What my intended plan was that I will estimate for each location, the distribution family and see if the parameter of that family are correlated with rainfall. In a way, I am trying to develop a model that links percentage of area planted with rainfall. If the parameters are related with rainfall, I can generate historical data of % planted every week using rainfall for past 10 years using the relationship I just established.
Any other tips to interpret this data and plots will be greatly appreciated