Posted: 11/12/2012 9:50:28 PM EDT
|
I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? |
|
Quoted:
I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? 87? |
|
Quoted:
Quoted:
I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? 87? That answer + your avatar = hilarious. |
|
Quoted:
I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? My statistical knowledge is restricted to insurance, but it seems to me like you could build a linear model by state and by category and do significance testing on the state factors. |
|
Quoted:
I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? If I could rephrase your question thusly: "Given a certain industry, are the HQs more likely to be in certain states than others?" -You could do a separate chi-squared analysis by industry. -You could do a t-test by industry and see if any of the states are outliers |
|
Quoted: Quoted: I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? If I could rephrase your question thusly: "Given a certain industry, are the HQs more likely to be in certain states than others?" -You could do a separate chi-squared analysis by industry. -You could do a t-test by industry and see if any of the states are outliers that's actually the direction i was thinking, but i wasn't sure how to derive expected frequencies for the individual chi-square GoF tests. i had also considered doing 1-sample Ts by state, but i was having trouble setting up the null. my question was sloppy––your formulation is much better. thanks!
|
|
Quoted:
Quoted:
Quoted:
I'm doing a chi-square contingency analysis on an 8x19 matrix. My result is huge, which is to be expected, with p = <0.001, phi = 0.91 and an effect size of 0.27. So obviously I have a very strong difference between f[o] and f[e] somewhere in the data. Now, I can eyeball where the major differences are, but I would really rather quantify them. So my question is this: is there some kind of post-hoc test that I can run to come up with significant difference thresholds for variable-on-variable? Was thinking about running individual goodness of fit tests, using the expected values derived from the contingency test, but I'm not sure if this would be mathematically valid. If it helps, the research question is based on 8 states and 19 categories of industry: are certain states preferred HQ locations for certain industrial sectors? Are there any resources in SPSS for this, or do i just need to eyeball it? If I could rephrase your question thusly: "Given a certain industry, are the HQs more likely to be in certain states than others?" -You could do a separate chi-squared analysis by industry. -You could do a t-test by industry and see if any of the states are outliers that's actually the direction i was thinking, but i wasn't sure how to derive expected frequencies for the individual chi-square GoF tests. i had also considered doing 1-sample Ts by state, but i was having trouble setting up the null. my question was sloppy––your formulation is much better. thanks! The expected frequency for the individual chi-squared tests would be the average frequency across that industry, wouldn't it? If your data looks like this: Industry 1 AL: 5 CA: 10 TX: 15 The expected would be 10 in each state. I'm pretty rusty on T-tests, but as a sort of "soft" test, I guess you could determine, say, the 95th %ile confidence interval for the frequency in each state, and then see what percentage of states actually fall outside the range and compare that percentage to 5%. |
|
Quoted: Quoted: that's actually the direction i was thinking, but i wasn't sure how to derive expected frequencies for the individual chi-square GoF tests. The expected frequency for the individual chi-squared tests would be the average frequency across that industry, wouldn't it? If your data looks like this: Industry 1 AL: 5 CA: 10 TX: 15 The expected would be 10 in each state. that makes sense. i'll need to sit down with my data tomorrow and confirm that that matches up with my derived expected freqs from the contingency test, because i'm going to get murdered if i present two different expected freqs. that's why i was thinking about going to the t-test, so that i could just present a t-score in results. but then i would have to compare a value to an average, instead of a mean to mu. thanks for taking the time with this. you might be amused to know that my professor's father wrote a "stats for insurance professionals" textbook.
|