WHERE_I_LEFT_OFF.txt

Did:

To Do:
1. Plot the population distributions, matplotlib
2. Compare with either python or R p-value calculators
4. Write up latex description of this program.
5. Make plots - check that hist is acutally normalized .... Needs fixed.
6. compare t-scores from both methods
7. Dynamically handle if stdev whether to use welch's or normal/paired t-test
8. Work on writing up notes in a concise fashion.
9. Compute effect size distribution when not homogenous
10. Break up code into several options that can be run separately

12. Given the info in Ch. 6 of practical meta-analysis, I don't understand how Smith and Iadarola (2015)
    computed confidence interval. Their CI should be symmetric, but they are not. 
    Revisit notes from meeting with Bill

13. Add README.md to outline notes/comments/thoughts in this file.
14. NOTE!!! the confidence intervals reported in Smith are not the same 
    confidence interals in Lipsey & Wilson. Their confidence intervals are fore the     
    combinatino of multipe studies _not_ individual studies.


Talked with Bill:
1. Talked to Bill, t-distribution is normal only when N-> large,
   For looking at N=3,5,10,20, should be very careful about computing
   frac_w_t_gt_2.  This corrects for the fact that at low N.  When N is small
   my p-value is wrong! Need to adjust for critical errors.
   --> He says 'more often than not' biomedical researcher does the right thing!
   --> 'unpaired' two sample t-test, '2-sample-independent' t-test.
   --> look up pt and qt
   --> FIXED!!
2. Suggested plotting 't' and comparing to normal
3. My stdev of mean intuition is correct.

Question:
1. Ask Bill if my random trials are now correct, now that I'm using the correct t-score?
   --> Ask about student-t vs. welch-t
   Answer : as 500 experiments -> 10**6, the frac_p<0.5 welch's and equal tscores should converge.  This is b/c I'm estimating s1 and s2 in welchs_t_score
2. Ask Bill why Cohen's d avoids sample size effects? \sigma_observed converges on \sigma_population and this is still a result of effect size