2
$\begingroup$

How one can compute the values of $\chi^2$-tables?

I saw two tables where was given for example that if degree of freedom is $1$ and $p=0.001$ then table value is in one table $10.827$ but in the another table $10.828$. I would like to know the formula which gives those values such that I can say which one is more accurate.

$\endgroup$
  • 1
    $\begingroup$ In what way does it matter to you? $\endgroup$ – Glen_b Mar 12 '15 at 23:19
  • $\begingroup$ I just like to know how the table values have been build. $\endgroup$ – self-studying Mar 12 '15 at 23:24
3
$\begingroup$

The $0.001$ upper tail critical value for a $\chi^2_1$ is $10.82756617...$

Since the chi-square with 1 df is the square of a normal, if you have more accuracy in z-tables you can square the 0.0005 upper tail critical value for a Z.

The "formula" for the tail area of a $\chi^2_1$ in effect requires the use of either the error function or the incomplete gamma function, which are generally evaluated by computer algorithms. For tail areas of chi-squares in general, the incomplete gamma function is used. For this problem (given the tail area, find the chi-squared value that gives it), you need the inverse incomplete gamma function.

The incomplete gamma function and its inverse function may be evaluated in a variety of ways -- for example, series approximation, through some iterative or recursive method, split into ranges where accurate functional approximations (possibly using ratios of polynomials for example) might be used. It's impossible to guess exactly what methods were used to produce your table.

I used a computer program to find the value to more digits - specifically, in R, I used

print(qchisq(.001,1,lower.tail=FALSE),d=12)
[1] 10.8275661707

(I think this is a couple more digits than the approximation will be accurate to.)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.