investigate categorizing the lines

[feomike]

• read and look at these
• http://stats.stackexchange.com/questions/27861/similarity-measures-between-curves
• http://quant.stackexchange.com/questions/848/time-series-similarity-measures
• https://www.google.com/webhp?sourceid=chrome-instant&ion=1&espv=2&ie=UTF-8#q=python%20frechet%20distance
• http://dataconomy.com/implementing-the-five-most-popular-similarity-measures-in-python/
• least squares - http://python4mpia.github.io/fitting_data/least-squares-fitting.html
• this one - http://stats.stackexchange.com/questions/19103/how-to-statistically-compare-two-time-series
• https://en.wikipedia.org/wiki/Granger_causality

[feomike]

so the current thinking is to investigate three statistics; (a) a difference signature (growing or declining) each year; (b) least squares line (linear more growing or declining over the time period); (c) granger or frechet distance to evaluate time series.

see these links

• http://stats.stackexchange.com/questions/27861/similarity-measures-between-curves
• https://en.wikipedia.org/wiki/Granger_causality
• https://www.google.com/webhp?sourceid=chrome-instant&ion=1&espv=2&ie=UTF-8#q=python+least+squares+regression+line

[feomike]

the key issue is deciding on the hypothesis first; that will illustrate how you choose a test; talk w/ alexei/alex