How would you guys go about working out the correlations between different time series of product prices; change price data into log returns, then enter into a correlation matrix? Or Arithmetic returns, or just use prices. Can’t remember exactly the differences between each, other than log returns can be added to give total return…and remember something vaguely about if using log returns then the data has to be normally distributed.

Or would you work out the difference between today’s price and yesterday’s price and then work out the returns of these numbers and correlations from that?

I know it's a lot of questions, if anything i can do the rest if i can just find out what returns i'm supposed to use here (or if i'm supposed to use prices), and how i can get the returns to produce something that makes sense with the negative/positive/0 values. ]]>

But I don't think that is common practice, again based on what I have read, and I am not sure if indeed you should do this. ]]>

I'm a fingerprint examiner by trade and got an idea after reading some papers on various quantitative methods of determining consensus and am seeking some input from those with experience. Fingerprint comparisons are a quantitative/qualitative field and are ultimately evaluative. Therefore, it should come as no surprise that the valuation of data in a fingerprint can vary from examiner to examiner. While this generally poses no problem, there can be times when the data can be valued in such a way that two examiners may come to different conclusions (Identification vs Inconclusive). These are disagreements of scale, not type (Identification vs Exclusion).

Since fingerprint conclusions are a factor of the data being used, the question I have then is, "Is ANOVA an appropriate detection method for honing in on what data is in question in a print?" Also, is my example the appropriate implementation of the method.

The method involves: alpha = .05 fyi

- Ask X amount of Fingerprint Examiners to chart corresponding data points between a latent print and a known print
- Combine all data into one chart and ask the same Examiners to rank the data 1 thru 5 based on a scale of objectivity
- Compile data into Excel and run two factor ANOVA w/o replication
- Look at the rows two see which data points are below the F-crit
- Use those data points as a basis of discussion on resolving conflict

Here's a sample image of what some imaginary data might look like:

Can someone comment on how they'd best make use of this type of data.

Thanks,

Boyd ]]>