Any Stats Gurus?

[AndyM]

Hi all,

A long shot but thought I'd send an appeal anyway. A friend of mine is completing a Dissertation where I've agreed to help out with some stats (I'm a Postgraduate Researcher in Cognitive Psychology), but I've not had to run stats like these before so am not entirely sure how to go about doing it.



The hypothesis is that despite no increase in intensity or frequency, damage costs from hurricanes are increasing (due to various factors such as increased coastal development), whilst lives lost have decreased (due to improved evacuation measures or whatever).

Variables are: Hurricane Strength (category rating), Damage Costs (USD adjusted for inflation), Lives Lost and Frequency, across Time (1960 - 2008).

My friend has run a number of Spearman's correlations between all these variables with Time. Now I'm not sure Time should be used as a variable in a correlation like this, and think an alternative test should be used - The question is which one?

I thought perhaps grouping the Time into decades, and then running an ANOVA with Time as the IV and mean damage costs, intensity and lives lost as a DV. However intensity is on a scale of 1-6 so I'm not sure if an ANOVA is suitable for this.

Alternatively I thought some sort of regression - Perhaps with Damage Costs as the outcome variable and the others as predictor variables (but can you include time as a predictor variable??)

Any thoughts would be welcome.

Cheers,
Andy

 

[Dax]

Time can't be a predictor as it's not TECHNICALLY a variable. Your first suggestion seems the most logical. Have you ran some testing with that idea on a handful of results to see how it looks? Obviously if it looks off you will need to ponder another method, but I can't contemplate a reason why it wouldn't work in a way you could use.

I'll have a ponder on this later and if I get a eureka moment I'll come back to you.

 

[AndyM]

Thanks Dax, appreciate it. We're going to have a look at the data later so I'll run the ANOVA and have a look.

Andy

 

[antisback]

Ask Jason

 

[AndyM]

I've already been grilling him about Bayesian modelling, but maybe I will

 

[Dax]

Hahaha I ****ing hate that ****.

 

[nopjes]

The bad part about the hurricane scale is that it does not include length, travelspeed, or anything, just "internal" wind speed, so in theory a F2 tornado could do more dmg than a F4 iirc, just by beeing active longer across a longer distance. So the active lenght of the tornado could be a usefull variable.

Alternatively, you could categorize the population density of a tornado's active area. Since in rural area's there is less to dmg. Or maybe, for example, communication in rural area's is worse resulting in a higher death rate.