Chi-Square Test: Solving Doubt w/ Kepler

[cptolemy]

Good afternoon,

I'm glad I've joined this forum. Here's my doubt: I have a serie of values in a table like this:

Case 1 34 55
Case 2 23 10
Case 3 55 40
etc...

the 34 means the observed value, and the 55 the control group, and so on. It's easy to do the test of course if...

The problem is: if the sum of the observed values is different from the sum of the control group, how do I execute the test?

Should I use %s and then, for instance, use a mean value from the sums...?

Kind regards,

Kepler

 

[I like Serena]

Good afternoon,

I'm glad I've joined this forum. Here's my doubt: I have a serie of values in a table like this:

Case 1 34 55
Case 2 23 10
Case 3 55 40
etc...

the 34 means the observed value, and the 55 the control group, and so on. It's easy to do the test of course if...

The problem is: if the sum of the observed values is different from the sum of the control group, how do I execute the test?

Should I use %s and then, for instance, use a mean value from the sums...?

Kind regards,

Kepler

Bi Kepler! Welcome to MHB! ;)

A chi square test only applies if we're talking about frequencies. That is, counts for some condition to occur.
That doesn't seem to be the case with your data. Can you clarify?
Otherwise a linear regression may be more appropriate...

 

[cptolemy]

Bi Kepler! Welcome to MHB! ;)

A chi square test only applies if we're talking about frequencies. That is, counts for some condition to occur.
That doesn't seem to be the case with your data. Can you clarify?
Otherwise a linear regression may be more appropriate...

Hi,

Thanks for the reply :) Actually they are frequencies where case 1,2,3... occurr. The control values are for regular and normal frequencies. The difference - and problem - is that the observed frequencies are being measured against a previous distribution - therefore the sums are different (the control cases where fewer). Chi square test relies on square differences. So I think I must choose the right proportion fo N.

I would very much like your opinion.

Kind regards,

Kepler

 

[I like Serena]

Hi,

Thanks for the reply :) Actually they are frequencies where case 1,2,3... occurr. The control values are for regular and normal frequencies. The difference - and problem - is that the observed frequencies are being measured against a previous distribution - therefore the sums are different (the control cases where fewer). Chi square test relies on square differences. So I think I must choose the right proportion fo N.

I would very much like your opinion.

Kind regards,

Kepler

A chi-square test typically compares observed frequencies against a hypothesized distribution.
Your control values are not a hypothesized distribution, but different observations of a group that is hypothesized to be different.

It means that a Kolmogorov-Smirnov test is more appropriate. It compares an observed distribution against a reference distribution, both with unknown distribution parameters.
Is it an option to use the Kolmogorov-Smirnov test?
Or does it have to be a chi-square test?

 

[cptolemy]

A chi-square test typically compares observed frequencies against a hypothesized distribution.
Your control values are not a hypothesized distribution, but different observations of a group that is hypothesized to be different.

It means that a Kolmogorov-Smirnov test is more appropriate. It compares an observed distribution against a reference distribution, both with unknown distribution parameters.
Is it an option to use the Kolmogorov-Smirnov test?
Or does it have to be a chi-square test?

Hi,

Thanks for the reply. Actually, you might be right. My observed values in a condition 1 belonging to a group of type A are compared to another group of the same (A) type without that condition. In condition 2, the observed group B is tested against another value of the same group without having condition 2 - and so on.

The difference - and problem - is that the obs. cases sum a sample of N individuals. The other group sums N1. N<>N1

But I must solve this for a chi square test.

Any help is apreciated.

Kind regards,

Kepler

 

[cptolemy]

A chi-square test typically compares observed frequencies against a hypothesized distribution.
Your control values are not a hypothesized distribution, but different observations of a group that is hypothesized to be different.

It means that a Kolmogorov-Smirnov test is more appropriate. It compares an observed distribution against a reference distribution, both with unknown distribution parameters.
Is it an option to use the Kolmogorov-Smirnov test?
Or does it have to be a chi-square test?

Hi,

Thanks for the reply. Actually, you might be right. My observed values in a condition 1 belonging to a group of type A are compared to another group of the same (A) type without that condition. In condition 2, the observed group B is tested against another value of the same group without having condition 2 - and so on.

The difference - and problem - is that the obs. cases sum a sample of N individuals. The other group sums N1. N is not equal to N1

Resume: I have several groups of type individuals, from A to F let's say. For a given condition, I have my observed values that comply with the condition (in a sample that sums N1 subjects) and a control value (the same type of group) but that does not complies that condition; and the subjects, N2, is different from N1.

But I must solve this for a chi square test.

Any help is apreciated.

Kind regards,

Kepler

 

[I like Serena]

Hi,

Thanks for the reply. Actually, you might be right. My observed values in a condition 1 belonging to a group of type A are compared to another group of the same (A) type without that condition. In condition 2, the observed group B is tested against another value of the same group without having condition 2 - and so on.

The difference - and problem - is that the obs. cases sum a sample of N individuals. The other group sums N1. N is not equal to N1

Resume: I have several groups of type individuals, from A to F let's say. For a given condition, I have my observed values that comply with the condition (in a sample that sums N1 subjects) and a control value (the same type of group) but that does not complies that condition; and the subjects, N2, is different from N1.

But I must solve this for a chi square test.

Any help is apreciated.

Kind regards,

Kepler

Is the observed group of type A the same as the observed group of type B?

If you really want to use a chi-square test, I think we will have to create a hypothesized distribution based on the control group.
We get that when we divide the observed frequency of the control group and divide it by the number of people in the control group. That gives us a proportion.
Then we can estimate the expected frequency by multiplying this proportion with the number of people in the observed group.
This approach is sensitive to errors in the measurements of the control group though, which would only be acceptable if the control group is very large.