oftentimes... i feel absolutely dumb.

i am not very good at getting what my supervisor here says on the first go; he has a rather distinct non-american accent, articulates fast and expects you to know what he is referring to like a picture in his head... that is when i do get a chance to actually have an approximately 10-20 minutes discussion... which is, by the way, rare. when anyone does get the chance... the conversations have to be really quick and to the point... results results remember? and often his advise or instructions are embedded into 2 or 3 succint sentences and i have to work out the code... you could literally say, it's all greek to me!

here are my bits of puzzles i have still to solve in order to get on to other things:

* i still can't figure out how to properly compute a standardized function of two coefficients from an autoregression analysis... (the linear model is for example: y = bo + bi*X1 + bii*X2 + e and i would like to related the two regression coefficients bi and bii into a single value but also standardize this joint value. if anyone is a stats guru out there could you help?!)

* i need the R-squared values from my autoregression but the model i am using, the exact maximum likelihood, does not output the R-sqs in the statistical programme (SPSS) that i am using... and i can't use the other models available even if they do output the R-sqs because they couldn't account for non-continuous time-series... i think the formula for R-sq is 1 minus the ratio (ResidualSumsOfSquaresOfErrors / TotalSumsOfSquaresOfErrors) and i think you can get the TotalSumsOfSquaresOfErrors from multiplying the Variance of the dependent variable by (the total number of samples(N) minus 1) so if the output gives the ResidualSumsOfSquaresOfErrors i could still calculate R-sq manually if I can also get the Variance of the dependent variable, perhaps... ?

** i wonder if the absence of the R-squared values in the output imply that it may not be used in the model to quantify the percentage of variance explained by the linear equation... hmm. if so, what might be it's equivalent index of measure for the 'goodness-of-fit'?

* i also have to work out the 'hypothetical' direction tuning curves using the regression coefficients...

oh my... to think i wake up everyday and feel like i don't know anything... i honestly do... and to add to that low morale, i have forgotten many things about stats from undergrad... or rather, i don't use it enough to know every little detail off the top of my head... and i have also forgotten much of my vector algebra and trigonometry from school... and can i note that matrix multiplication and transformation can be a pain? ... i used to be able to do them without too much hesitation, but that was many many many years ago and often you do them in context of specified mathematical problems... why don't they ever teach you real applied mathematics and science so you can appreciate its usefulness from day one and still remember how to do it some 15 years later?! like in regression equations?

*sob*

posted by ~overacuppa~ on Saturday, 25 March, 2006 at 00:14 hrs