J
jensen
Member
Hi
In Chapter 12, the example applying ChiSq test to graduation A and B (page 30) uses the test statistic of the form:
Sum {(Actual - Expected)^2}/Expected.
In Chapter 13 (section 5.2 pg 23), to test the fit of graduated rates the test statistic there is:
Sum {standardised deviations, z_i}^2
There is no difference between the two above if the Poisson model was used, but not so for the Binomial model since the variance have an additional (1-q) term. One can argue that q is approx 0, hence negligible, but it bugs me that the notes are not consistent.
In the exam, can we assume the q is approx 0, hence the first form of ChiSq test statistic applies?
In Chapter 12, the example applying ChiSq test to graduation A and B (page 30) uses the test statistic of the form:
Sum {(Actual - Expected)^2}/Expected.
In Chapter 13 (section 5.2 pg 23), to test the fit of graduated rates the test statistic there is:
Sum {standardised deviations, z_i}^2
There is no difference between the two above if the Poisson model was used, but not so for the Binomial model since the variance have an additional (1-q) term. One can argue that q is approx 0, hence negligible, but it bugs me that the notes are not consistent.
In the exam, can we assume the q is approx 0, hence the first form of ChiSq test statistic applies?