Dear all,
I would like to look to use bootstrap to look at the difference between the ratio outcome of two groups (Ratio group 1 = 0.36 VS Ratio group 2 = 0.29). See attached my syntax for calculating the ratio for each group and the outcome of this syntax in my SPSS data file. Syntax_DataSet1.sps <http://spssx-discussion.1045642.n5.nabble.com/file/t341365/Syntax_DataSet1.sps> DataSet1.sav <http://spssx-discussion.1045642.n5.nabble.com/file/t341365/DataSet1.sav> Because the outcome of the syntax is just one value for each group (the ratio is an outcome of the mean of one variable divided by the mean of another) there is no distribution of the ratio in each group and so a statistic analysis to determine the statistical difference between these groups is not possible. That’s why I thought about using bootstrapping. In this case I would request 1000 bootstrap samples and then perform my analyses (syntax) for each sample. My outcome would then consist of 1000 ratio’s (from the syntax analysis of 1000 samples) for each group and then I would be able to compare these median’s between the groups. Am I right about this assumption? And is it possible to perform this kind of bootstrapping? Does anyone know how to perform this or how to write these steps in my syntax? Thanks in advance!! With kind regards, JM -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
You do know the distribution of the means that make up the ratios - they are presumed to be Normal, with their computable means and variances.
There is a formula to get the variance of a function of two variables. You can apply it to your ratios if the denominators are far from zero. Thus, you /can/ compute error terms for your ratios, given the means and covariance matrices.
However, your denominator variable has "residual" in its name, which screams a warning here,
that your variance might be infinite. (I did not open with the warning, because your reported
ratios are small, being less than 1, so it does not look like "residual" is zero-centered.) That is: As it happens, the easy way to generate a "Cauchy" distribution, which has an infinite variance and undefined mean, is to take the ratio of two normal distributions. - If your "residuals" include zero in their range, ratios might be very unstable, that is,
reflect their "undefined" means.
If you describe what your variables are and what you hope to achieve, someone might
have other suggestions. I can't read your data, so I don't know what it shows. To me,
the situation sounds like it /might/ lend itself to monte-carlo simulation, randomly assigning cases to two samples 10,000 times, better than it does to bootstrap.
-- Rich Ulrich From: SPSSX(r) Discussion <[hidden email]> on behalf of Juleke2 <[hidden email]>
Sent: Tuesday, December 19, 2017 2:48:22 PM To: [hidden email] Subject: Bootstrap in syntax Dear all,
I would like to look to use bootstrap to look at the difference between the ratio outcome of two groups (Ratio group 1 = 0.36 VS Ratio group 2 = 0.29). See attached my syntax for calculating the ratio for each group and the outcome of this syntax in my SPSS data file. Syntax_DataSet1.sps <http://spssx-discussion.1045642.n5.nabble.com/file/t341365/Syntax_DataSet1.sps> DataSet1.sav <http://spssx-discussion.1045642.n5.nabble.com/file/t341365/DataSet1.sav> Because the outcome of the syntax is just one value for each group (the ratio is an outcome of the mean of one variable divided by the mean of another) there is no distribution of the ratio in each group and so a statistic analysis to determine the statistical difference between these groups is not possible. That’s why I thought about using bootstrapping. In this case I would request 1000 bootstrap samples and then perform my analyses (syntax) for each sample. My outcome would then consist of 1000 ratio’s (from the syntax analysis of 1000 samples) for each group and then I would be able to compare these median’s between the groups. Am I right about this assumption? And is it possible to perform this kind of bootstrapping? Does anyone know how to perform this or how to write these steps in my syntax? Thanks in advance!! With kind regards, JM -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Free forum by Nabble | Edit this page |