quick chi-square question

y2jake215

If I'm calculating chi-square for a set of data, and the expected values for one of the categories is 0, do I include that category as a degree of freedom?

For example, I'm calculating the chi-square for a data set of observed phenotypes in a F2 generation of flies. The expected value of 2 of the 8 possible outcomes, which have white-eyed females, are 0. We didn't find any, but what would I do if I HAD found a white-eyed female? I'd have to divide by zero. This is the one part of chi-square I realized I don't know what I would have to do, and I'd like to fully understand it.

ceres

I don't have advice because I start with chi squares in my genetics class on Monday. But I feel your pain and I'm interested in the answer.

At worst, maybe I can help on Monday.

Daenris

I'm fairly certain that the assumptions of a chi-square test don't hold if any of the expected outcomes are 0 (or even if many of the outcomes are very low for that matter).

Are you sure you should be using a chi-square test here? Should you maybe be using Fisher's Exact test or something else?

DJ-99

Are the only possible values 0 or greater? If so, how could you possibly have an EV of 0 but then have a value greater than 0?

If it's possible to have a value greater than 0, then the EV can't be 0. So you shouldn't ever have to divide by 0.

Unless it's possible to have negative values here, but I'm not really sure what your data is.

y2jake215

What we did was take a male fly with a white-eyed mutation and a female with a vestigial wing mutation and bred them, getting a heterozygous wild-type F1 generation, then self crossed to get an F2 generation. The F2 generation had all sorts of combinations of the two mutations and gender in certain ratios, but since white-eyes is a sex-linked trait, it doesn't show up in any females. The problem is, the TA gave us data to show what the exact amount of each fly expected is for a total of 1000 flies. His data has 0s next to the white-eyed female.

Female Male
Wild-type 375 187
White eyes, wild-type wings 0 187
Vestigial wings, wild-type eyes 125 63
Vestigial wings and white eyes 0 63
Total 500 500

I figure I should just ignore the 2 0s when calculating chi-square, but when getting the p-value I'm still unclear whether to use them as degrees of freedom (it still IS possible to get a white-eyed female, it just would take a lot of unlikely events)

Gdiguy

So those are the expected values - do you actually have white eyes in the females?

If your expected value is zero, and you don't get any events, then the chi-square test is valid - you should just ignore those two as they're not possible outcomes (so the degrees of freedom are 2 less). The chi-square test only tests against possible outcomes; if you know an outcome isn't possible, you don't need to include it. (Which makes sense - if you were doing a chi-square test after a homozygous AA x heterozygous Aa cross, you wouldn't include an outcome for an aa outcome - you know it's not possible, so you wouldn't include it).


If your expected value is zero and you DO get events, then you shouldn't be using a chi-square test for that data - if the expected value is close to zero, you can do tricks to get some reasonable approximate chi-square value, but if it's exactly zero the test breaks down (or really, balloons to infinity when expected values approach zero, which works the way it should - if you cannot possibly get an outcome with your null hypothesis, but you see that outcome, then your null hypothesis should be rejected)

Alistair Hutton

Daenris wrote: »

I'm fairly certain that the assumptions of a chi-square test don't hold if any of the expected outcomes are 0 (or even if many of the outcomes are very low for that matter).

Are you sure you should be using a chi-square test here? Should you maybe be using Fisher's Exact test or something else?

If any value is 5 or less (and some people would baulk at 10 or less) then the Chi-square does not hold. You are correct, the Fisher's Exact test should be use. There's used to be a really good one online but I can't seem to find it, the Wikipedia page has links to several ones.

apeelewuetu

I am presently working on a project in which i want to calculate chi - square , i want u to help me do the calculation, so that i can compare my own with your answer- Items 22 23
Response YES NO YES NO
Percentage after counselling 80 20 78 22
X2 =

apeelewuetu

Items 22 23
Response YES NO YES NO
Percentage after counselling 80 20 78 22
X2 133.68
is this value correct ?