search

mkao006 / sws_imputation

Repository for implementing imputation for the Statistical Working System
0 stars 0 forks source link

Quantile or Cauchy distribution #10

Closed mkao006 closed 11 years ago

mkao006 commented 11 years ago

Should we use quantile bounds or 95% confidence interval of the Cauchy distribution to validate our growth rate?

The quantile appears to be better at capturing the skewness, but the Cauchy is better at capturing the tail. Otherwise, we can just use the range to validate the growth.

mkao006 commented 11 years ago

Quantile has the advantage of capturing the skewness, we can use 25% and 75% plus minus the distance from the 50%.

Also test what the 3 standard deviation bound as well.

mkao006 commented 11 years ago

The modified quantile range (first quantile - median - (first quantile) and 3rd quantile + (3rd quantile - median)) seems to work better than the Cauchy and the 3 standard deviation bound.

Lets have a look at country specific changes to see if there is a major change in the distribution, and whether there is sufficient data to support country specific validation.

mkao006 commented 11 years ago

Investigate why some plots can not be generated, is there in sufficient data or estimation?

mkao006 commented 11 years ago

Issue continued in #21