Printable View
I'd suggest coding the visibility reports as categorical response data and reformulate your hypothesis as a falsifiable null. Then, if the environmental data, coded as multivariate predictors, yield significant effects in, say, a MANOVA, construct a set of discriminate functions to make predictions and use cross-validation to gauge it effectiveness. Child's play. The only REAL effort involves aggregating the data...
Dave
That makes sense. How many categories would you use for the observation reports? 5 (e.g. terrible, poor, OK, good, excellent)?
What is the null hypothesis? Is it "environmental condition x does not predict visibility result"?
If I put it all in an Excel spreadsheet and send it to you, can you do the rest? I have only a very basic understanding of statistics.
Yes, I think a categorical visibility variable with 5 levels would work well.
Yes, the Ho: "No significant effect of environmental variable(s) on visibility". We can use a variable selection procedure to construct a parsimonious model, which only includes the environmental variables important in predicting visibility.
If you can aggregate the data, I'm happy to do the analysis. At table in Excel would be great, with rows = observations and columns = variables. One column should be date, so we can examine temporal lags in the response of visibility to the environmental variables over a range of time scales. I'll PM you my email.
Dave