2
$\begingroup$

I have a question on testing mortality table. Suppose I am given a simple mortality table:

age  | prob of dying  | prob of surviving
---------------------------------------
20   |  0.01          |   0.99
21   |  0.02          |   0.98
22   |  0.03          |   0.97
23   |  0.04          |   0.96
...

I want to test whether the table fit my observed data, i.e., whether the Observed values

age  | actual dead    | actual survivors
---------------------------------------
20   | 0              |  397
21   | 1              |  189
22   | 0              |  136
23   | 2              |  100
...

fit the expected values

age  | actaul dead       | actual survivors
---------------------------------------
20   | 3.97 (397*0,01)   |  393.03
21   | 3.8   (190*0.02)  |  186.2
22   | 4.08  (136*0.03)  |  131.92
23   | 4.08 (102*0.04)   |  97.92
...

How do I do that? What method should I use? Should I use Chi-Square, although my expected data is quite small? or there is another method?

$\endgroup$
1
$\begingroup$

The expecteds aren't all that low, the chi-squared should work fine.

There are other alternatives.

You could do an "exact" test, akin to a Fisher Exact test, but based off the binomial.

You could even do a G-test, another large sample test based off the likelihood-ratio test.

$\endgroup$
  • $\begingroup$ Thank you very much for the answer. But can I use Fisher's Exact test in this situation, I thought it is used as a test of independence? in my example the expected values are not calculated by the formula $\endgroup$ – naninko Apr 16 '13 at 12:30
  • $\begingroup$ Hence the use of the phrase "akin to"; it's not actually the exact test that Fisher did. I made some changes to what I wrote; the 'condition on both margins' didn't make sense. $\endgroup$ – Glen_b -Reinstate Monica Apr 16 '13 at 12:40
  • $\begingroup$ Thank you very much for the answer. I alreeady used Chi Square and G test. I am not sure if that is the right apporach, since I already have predetermined probabilities, so my expectations are calculated by the formula $$ E(X_{20})=p_{20}*n_{20} $$ and not by the formula used in exact tests for contingcency tables (row*column/total_number) $$ E(X_{20})=\frac{n_{20}*d}{n} $$ $p_{20}$ -prob a 20 year old to die $n_{20}$ -total number of 20 year olds $d$- total number of dead $Could I still use a test of independence, like Fisher's or chi square, or should I consider a different approach? $\endgroup$ – naninko Apr 16 '13 at 13:04
  • $\begingroup$ Check the first paragraph here, and then this section $\endgroup$ – Glen_b -Reinstate Monica Apr 16 '13 at 13:40

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.