complex samples (stratified sampling)
Locked
complex samples (stratified sampling)
 
|
Hello SPSS experts!
I'm not sure if this is more of SPSS question of more of a statistical design question, but I'm hoping someone can give me some advice. I'm using the "complex samples" feature to take a sample of 65 records from a universe of 146 records. I'm stratifying on 3 variables. One variable is poverty which is two options (above or below poverty level), one variable is geography (urban, suburban, and rural), and one field is distance (which is a yes/ no question). So I set up the counts that I'd like to take within each combination of elements (12 possibilities) and used those counts to draw the sample (we are oversampling on rural and using the population proportions for the rest of the sample).
My problem is that there are some missing values for each of the fields. So instead of getting sample back of 65, I'm getting 55 back. I think the missing values on the strata is causing the sample to not return the requested counts.
Is there a way deal with missing values on a strata in SPSS? Is there some way I should calculate the counts in a strata where one of the fields is missing?
Hopefully my question makes sense.
Thanks for any help or advice you can provide!
Matt
-- Matthew Hoover
Research Associate Center for Youth and Communities Heller School for Social Policy and Management MS035 Brandeis University 415 South Street Waltham, MA 02454 TEL 781.736.8631 FAX 781.736.3773 |
|
Complex sample is to select a random sample from extra large data say like if you had 146000 or 146000000 records…not 146 records with all the missing values you mention J Max. From: Matthew Hoover [via SPSSX Discussion] [mailto:[hidden email]] I'm not sure if this is more of SPSS question of more of a statistical design question, but I'm hoping someone can give me some advice. I'm using the "complex samples" feature to take a sample of 65 records from a universe of 146 records. I'm stratifying on 3 variables. One variable is poverty which is two options (above or below poverty level), one variable is geography (urban, suburban, and rural), and one field is distance (which is a yes/ no question). So I set up the counts that I'd like to take within each combination of elements (12 possibilities) and used those counts to draw the sample (we are oversampling on rural and using the population proportions for the rest of the sample).  My problem is that there are some missing values for each of the fields. So instead of getting sample back of 65, I'm getting 55 back. I think the missing values on the strata is causing the sample to not return the requested counts.  Is there a way deal with missing values on a strata in SPSS? Is there some way I should calculate the counts in a strata where one of the fields is missing?  Hopefully my question makes sense.  Thanks for any help or advice you can provide!  Matt |
Re: complex samples (stratified sampling)
|
|
We did a recruitment survey online and from those responses we are taking a random sample that will be selected to be in the larger study. We can't use everyone who responded (we just need around 50 - 60 per group).
FYI, I'm using v19 if that makes a difference for my question.
Matt
On Wed, Sep 19, 2012 at 1:31 PM, Max Jasper <[hidden email]> wrote:
-- Matthew Hoover
Research Associate Center for Youth and Communities Heller School for Social Policy and Management MS035 Brandeis University 415 South Street Waltham, MA 02454 TEL 781.736.8631 FAX 781.736.3773 |
|
|
Why not taking your 65 selections from the original large survey data? Instead of reducing it to 146 then selecting 65? Max. From: Matthew Hoover [mailto:[hidden email]] We did a recruitment survey online and from those responses we are taking a random sample that will be selected to be in the larger study. We can't use everyone who responded (we just need around 50 - 60 per group). FYI, I'm using v19 if that makes a difference for my question. Matt On Wed, Sep 19, 2012 at 1:31 PM, Max Jasper <[hidden email]> wrote: Complex sample is to select a random sample from extra large data say like if you had 146000 or 146000000 records…not 146 records with all the missing values you mention J Max. From: Matthew Hoover [via SPSSX Discussion] [mailto:[hidden email]] I'm not sure if this is more of SPSS question of more of a statistical design question, but I'm hoping someone can give me some advice. I'm using the "complex samples" feature to take a sample of 65 records from a universe of 146 records. I'm stratifying on 3 variables. One variable is poverty which is two options (above or below poverty level), one variable is geography (urban, suburban, and rural), and one field is distance (which is a yes/ no question). So I set up the counts that I'd like to take within each combination of elements (12 possibilities) and used those counts to draw the sample (we are oversampling on rural and using the population proportions for the rest of the sample).  My problem is that there are some missing values for each of the fields. So instead of getting sample back of 65, I'm getting 55 back. I think the missing values on the strata is causing the sample to not return the requested counts.  Is there a way deal with missing values on a strata in SPSS? Is there some way I should calculate the counts in a strata where one of the fields is missing?  Hopefully my question makes sense.  Thanks for any help or advice you can provide!  Matt
Matthew Hoover |
Re: complex samples (stratified sampling)
|
|
In reply to this post by MaxJasper
"Complex samples" are for LARGE samples. Yes, that was my impression, too.
Stratifying creates a bias exactly in the instances where it is most tempting to amateurs -- where the proportions above/below a cutoff are different. That's one reason to avoid what is proposed. And it is usually a bad idea to throw away data. PS. the stuff with the missing data must cause the wrong Ns. But fixing some erroneous code is especially hard without seeing the code. -- Rich Ulrich Date: Wed, 19 Sep 2012 11:31:15 -0600 From: [hidden email] Subject: Re: complex samples (stratified sampling) To: [hidden email] Complex sample is to select a random sample from extra large data say like if you had 146000 or 146000000 records…not 146 records with all the missing values you mention J
Max.
From: Matthew Hoover [via SPSSX Discussion] [mailto:[hidden email]] I'm not sure if this is more of SPSS question of more of a statistical design question, but I'm hoping someone can give me some advice. I'm using the "complex samples" feature to take a sample of 65 records from a universe of 146 records. I'm stratifying on 3 variables. One variable is poverty which is two options (above or below poverty level), one variable is geography (urban, suburban, and rural), and one field is distance (which is a yes/ no question). So I set up the counts that I'd like to take within each combination of elements (12 possibilities) and used those counts to draw the sample (we are oversampling on rural and using the population proportions for the rest of the sample).  My problem is that there are some missing values for each of the fields. So instead of getting sample back of 65, I'm getting 55 back. I think the missing values on the strata is causing the sample to not return the requested counts.  Is there a way deal with missing values on a strata in SPSS? Is there some way I should calculate the counts in a strata where one of the fields is missing?  Hopefully my question makes sense.  Thanks for any help or advice you can provide!  Matt |
Re: complex samples (stratified sampling)
|
|
In reply to this post by Matthew Hoover
Your online recruitment survey is itself a form of self-selection and thus its results are not a random sample. The whole theory of sampling is based on the so-called Law of Large Numbers, which refers to LARGE samples that are drawn at RANDOM. Applying stratified sampling and Complex Samples to your 146 self-recruited cases is just overkill, and would give a misplaced impression of scientific precision about a procedure that is inherently non scientific. Hector De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de Matthew Hoover We did a recruitment survey online and from those responses we are taking a random sample that will be selected to be in the larger study. We can't use everyone who responded (we just need around 50 - 60 per group). FYI, I'm using v19 if that makes a difference for my question. Matt On Wed, Sep 19, 2012 at 1:31 PM, Max Jasper <[hidden email]> wrote: Complex sample is to select a random sample from extra large data say like if you had 146000 or 146000000 records…not 146 records with all the missing values you mention J Max. From: Matthew Hoover [via SPSSX Discussion] [mailto:[hidden email]] I'm not sure if this is more of SPSS question of more of a statistical design question, but I'm hoping someone can give me some advice. I'm using the "complex samples" feature to take a sample of 65 records from a universe of 146 records. I'm stratifying on 3 variables. One variable is poverty which is two options (above or below poverty level), one variable is geography (urban, suburban, and rural), and one field is distance (which is a yes/ no question). So I set up the counts that I'd like to take within each combination of elements (12 possibilities) and used those counts to draw the sample (we are oversampling on rural and using the population proportions for the rest of the sample).  My problem is that there are some missing values for each of the fields. So instead of getting sample back of 65, I'm getting 55 back. I think the missing values on the strata is causing the sample to not return the requested counts.  Is there a way deal with missing values on a strata in SPSS? Is there some way I should calculate the counts in a strata where one of the fields is missing?  Hopefully my question makes sense.  Thanks for any help or advice you can provide!  Matt
Matthew Hoover |
| Free forum by Nabble | Edit this page |