1

I currently try to transform a Gamma into a Chi-square (X2) and calculate a 95% confidence interval. So, let's say I have a Gamma distribution of $$Gamma(4.5, 4)$$
Given the answer in https://math.stackexchange.com/questions/1575050/transforming-gamma-into-chi-squared-distribution I can transform the gamma into a chi-square by

enter image description here

To get theta, I do

$$\frac{1}{2\theta}=4$$ $$\theta = \frac{1}{8}$$

Then, I would get

$$8Y = \chi^2(2*4.5) = \chi^2(9)$$

I can get the constants from a X2-table for 9 degrees of freedom with

$$lower = 2.70$$

$$upper = 19.02$$

I would then multiply by 8 to get $$P(8*lower < Y < 8*upper)$$ for the desired interval for the gamma, which is way too large.

However, when I use R to calculate the upper and lower bounds by using

alpha <- 4.5 
beta <- 4

alpha_conf <- 0.05
q_lower <- qgamma(alpha_conf / 2, alpha, beta)
q_upper <- qgamma(1 - alpha_conf / 2, alpha, beta)
c(q_lower, q_upper)

I get $$P(0.3375487 < Y < 2.3778460)$$ which makes much more sense looking at the data. I realized that the constants I got from the X2-table will give me the same result, if I do $$P(\frac{1}{8}*lower < Y < \frac{1}{8}*upper)$$

If I use the formula given in Probability Interval for Gamma Distribution I get yet another solution (alpha and beta are reverse to my syntax, so in that case it would be Gamma(4, 4.5)). With

$$\frac{2*Y}{4} \sim \chi^2(9)$$ $$\frac{1}{2}*Y \sim \chi^2(9)$$,

which, again, would not give me the correct result.

Assuming that the calculations in R are correct, I don't know where my mistake is. Thanks for the help!

noog4
  • 23
  • 1
    Please see https://math.meta.stackexchange.com/questions/5020/mathjax-basic-tutorial-and-quick-reference. – StubbornAtom Feb 10 '20 at 19:34
  • sure, if that is the only problem that's an easy fix! – noog4 Feb 10 '20 at 20:19
  • Just figured it out myself. 8*lower = 2.7 => lower = 2.7/8 would be the the lower bound of the original gamma. Simmilarly, for the upper. – noog4 Feb 10 '20 at 22:39

0 Answers0