SPSS Statistics

How to measure under/overrepresentation of results in SPSS?

I run an employee survey. I want to assess:

1. to what extent the results reflect the population demographics, overall (is there a single indicator?)
2. which groups are significantly overrepresented or underrepresented in the results

I know the population demographics (= how many employees are in each subgroup), and I also have the responses (=sample) by subgroup. 

So, I can easily calculate what the response rate (percentage) is by subgroup and overall. 

So, I can also run a Chi-Square analysis and compare within each demographic category if certain profiles have significantly higher response rates than other, for example, do certain functions respond more than other functions, do people with low seniority respond less frequently than people with longer tenure etc.

The question:

But how can I quantify to what extent certain groups (functions, seniority levels, …) are overrepresented or underrepresented in the survey results as a whole?

Imagine visually two pie charts. One with the population distribution over functions, seniority levels, mgr/employee, location… all in one pie chart. And another one with the distribution of the same demographics in the sample (the responses). I guess, I need to compare the proportions of the population pie with the proportions of the sample pie. 

I have difficulty putting that into practice in SPSS. How to organize my datasheet? Do I need to use an intermediate step through OMS? What analysis to choose? Any ideas?

A related question could be: is there something like "post-stratification" possible in SPSS? In other words, if groups are underrepresented or overrepresented, can I extrapolate / reduce their results to make them reflect the population more accurately?

How to measure under/overrepresentation of results in SPSS?

I run an employee survey. I want to assess:

1. to what extent the results reflect the population demographics, overall (is there a single indicator?)
2. which groups are significantly overrepresented or underrepresented in the results

I know the population demographics (= how many employees are in each subgroup), and I also have the responses (=sample) by subgroup. 

So, I can easily calculate what the response rate (percentage) is by subgroup and overall. 

So, I can also run a Chi-Square analysis and compare within each demographic category if certain profiles have significantly higher response rates than other, for example, do certain functions respond more than other functions, do people with low seniority respond less frequently than people with longer tenure etc.

The question:

But how can I quantify to what extent certain groups (functions, seniority levels, …) are overrepresented or underrepresented in the survey results as a whole?

Imagine visually two pie charts. One with the population distribution over functions, seniority levels, mgr/employee, location… all in one pie chart. And another one with the distribution of the same demographics in the sample (the responses). I guess, I need to compare the proportions of the population pie with the proportions of the sample pie. 

I have difficulty putting that into practice in SPSS. How to organize my datasheet? Do I need to use an intermediate step through OMS? What analysis to choose? Any ideas?

A related question could be: is there something like "post-stratification" possible in SPSS? In other words, if groups are underrepresented or overrepresented, can I extrapolate / reduce their results to make them reflect the population more accurately?

Thank you very much, Jon. That has helped me a lot. I discovered along the way that you are actually an author of a paper on the subject!

Let me put down a few of my learnings for the benefit of whoever is looking at doing the same. And, Jon, correct me if I state errors below.

• When specifying the variable values, that is done in an SPSS numerical way (f.i. 10012), syntax would read for example: SPSSINC RAKE DIM1 = var3 10001 9 10002 71 (sample has 9 managers and 71 individual contributors in this case)
• It is not necessary to separately calculate the proportions (percentages) in the population [which I had done at first], putting in the numbers for each value is sufficient. SPSS will calculate based on the total population figure.  See example above.
• The "Sample Balance" which is obtained indicates how well the weighted sample matches the population proportions, 100 being a perfect match. Additional question: how can I find out how far the unweighted sample was off?  It would be nice to know that the weighting moved the needle from 60 to 92, for example.
• True, the heatmap and raked weights table give an indication of how strongly certain variables had to be overweighted or underweighted to produce the new balanced sample. But is there a single number that is comparable to the balanced sample? Something like an unbalanced sample?
• I have not yet figured out exactly how the PANELDOWNVAR=varname PANELACROSS=varname work precisely for plotting the heatmap, but that is a matter of trying out.
• To put weighting off again, there are two ways. The brutal one is to delete (=clear) the weighting variable from the dataset. The other way is to go to Data – Weight Cases [not Rake Weights] – and select Do not weight cases – OK.

Thanks again and open to further remarks.

How to measure under/overrepresentation of results in SPSS?

I run an employee survey. I want to assess:

1. to what extent the results reflect the population demographics, overall (is there a single indicator?)
2. which groups are significantly overrepresented or underrepresented in the results

I know the population demographics (= how many employees are in each subgroup), and I also have the responses (=sample) by subgroup. 

So, I can easily calculate what the response rate (percentage) is by subgroup and overall. 

So, I can also run a Chi-Square analysis and compare within each demographic category if certain profiles have significantly higher response rates than other, for example, do certain functions respond more than other functions, do people with low seniority respond less frequently than people with longer tenure etc.

The question:

But how can I quantify to what extent certain groups (functions, seniority levels, …) are overrepresented or underrepresented in the survey results as a whole?

Imagine visually two pie charts. One with the population distribution over functions, seniority levels, mgr/employee, location… all in one pie chart. And another one with the distribution of the same demographics in the sample (the responses). I guess, I need to compare the proportions of the population pie with the proportions of the sample pie. 

I have difficulty putting that into practice in SPSS. How to organize my datasheet? Do I need to use an intermediate step through OMS? What analysis to choose? Any ideas?

A related question could be: is there something like "post-stratification" possible in SPSS? In other words, if groups are underrepresented or overrepresented, can I extrapolate / reduce their results to make them reflect the population more accurately?

Thank you very much, Jon. That has helped me a lot. I discovered along the way that you are actually an author of a paper on the subject!

Let me put down a few of my learnings for the benefit of whoever is looking at doing the same. And, Jon, correct me if I state errors below.

• When specifying the variable values, that is done in an SPSS numerical way (f.i. 10012), syntax would read for example: SPSSINC RAKE DIM1 = var3 10001 9 10002 71 (sample has 9 managers and 71 individual contributors in this case)
• It is not necessary to separately calculate the proportions (percentages) in the population [which I had done at first], putting in the numbers for each value is sufficient. SPSS will calculate based on the total population figure.  See example above.
• The "Sample Balance" which is obtained indicates how well the weighted sample matches the population proportions, 100 being a perfect match. Additional question: how can I find out how far the unweighted sample was off?  It would be nice to know that the weighting moved the needle from 60 to 92, for example.
• True, the heatmap and raked weights table give an indication of how strongly certain variables had to be overweighted or underweighted to produce the new balanced sample. But is there a single number that is comparable to the balanced sample? Something like an unbalanced sample?
• I have not yet figured out exactly how the PANELDOWNVAR=varname PANELACROSS=varname work precisely for plotting the heatmap, but that is a matter of trying out.
• To put weighting off again, there are two ways. The brutal one is to delete (=clear) the weighting variable from the dataset. The other way is to go to Data – Weight Cases [not Rake Weights] – and select Do not weight cases – OK.

Thanks again and open to further remarks.

How to measure under/overrepresentation of results in SPSS?

I run an employee survey. I want to assess:

1. to what extent the results reflect the population demographics, overall (is there a single indicator?)
2. which groups are significantly overrepresented or underrepresented in the results

I know the population demographics (= how many employees are in each subgroup), and I also have the responses (=sample) by subgroup. 

So, I can easily calculate what the response rate (percentage) is by subgroup and overall. 

So, I can also run a Chi-Square analysis and compare within each demographic category if certain profiles have significantly higher response rates than other, for example, do certain functions respond more than other functions, do people with low seniority respond less frequently than people with longer tenure etc.

The question:

But how can I quantify to what extent certain groups (functions, seniority levels, …) are overrepresented or underrepresented in the survey results as a whole?

Imagine visually two pie charts. One with the population distribution over functions, seniority levels, mgr/employee, location… all in one pie chart. And another one with the distribution of the same demographics in the sample (the responses). I guess, I need to compare the proportions of the population pie with the proportions of the sample pie. 

I have difficulty putting that into practice in SPSS. How to organize my datasheet? Do I need to use an intermediate step through OMS? What analysis to choose? Any ideas?

A related question could be: is there something like "post-stratification" possible in SPSS? In other words, if groups are underrepresented or overrepresented, can I extrapolate / reduce their results to make them reflect the population more accurately?