Hello everyone. I am using a big database of more than 3 million observations and around 60 variables. It contains longitudinal data from the early 80s to late 10s.
Subjects are not individual people, but rather each job contract, identified by a personal identification number. This is done so since sometimes the same person can have two or more job contracts at the same time (not really frequent, but also not unheard of). Each subject/contract has among others, two date variables: start date and end date of said contract.
I am interested in observing the difference in duration between contracts during two time periods: 2005 to 2010 (period A) and 2010 to 2015 (period B).
Using those two start and end date variables and the help of a kind user of the forum, I was able to create the duration variable for each job contract, and also two extra variables, duration for the job contract during period A (durA) and duration for the job contract during period B (durB). You can check the original question here: https://www.statalist.org/forums/for...=1685987788624
That means there are many missing data in those latter two variables (as not all the contracts span the duration of those two periods, for example a job contract contained in the database might have started in 1990 and ended in 1995, thus in durA and durB it would have only missing data, this would be scenario 1). Some contracts might have only values for durA (eg. a contract that begun in 2006 and ended in 2009) or for durB (eg. a contract that lasted from 2011 to 2013; this would be scenario 2). And finally (scenario 3), some contracts might have data for both durA and durB (a contract that began in 2000 and lasted until 2016 would have the maximum value for both durA and durB).
As far as I know, I can't perform a t-test on those two variables, as the data are only partially paired (those who fit into scenario 3). Many contracts only have values for either durA or durB. I've read a few papers on techniques to make comparisons involving weighted t-tests with partially paired data (eg. https://www.tqmp.org/RegularArticles.../p055/p055.pdf, https://digitalcommons.wayne.edu/cgi...&context=jmasm, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9041716/, https://citeseerx.ist.psu.edu/viewdo...=rep1&type=pdf), but there is no explanation on how to implement them on Stata (one of the papers mentions a package for R, though).
Would Welch's t-test work in this case? Or is there a way to perform the analysis using the main duration variable (the one that includes the duration of all contracts) but comparing the duration of only those two periods (that is, the values included between on one hand 2005-2010 and 2010-2015 on the other).
Thank you so much for your help.
Subjects are not individual people, but rather each job contract, identified by a personal identification number. This is done so since sometimes the same person can have two or more job contracts at the same time (not really frequent, but also not unheard of). Each subject/contract has among others, two date variables: start date and end date of said contract.
I am interested in observing the difference in duration between contracts during two time periods: 2005 to 2010 (period A) and 2010 to 2015 (period B).
Using those two start and end date variables and the help of a kind user of the forum, I was able to create the duration variable for each job contract, and also two extra variables, duration for the job contract during period A (durA) and duration for the job contract during period B (durB). You can check the original question here: https://www.statalist.org/forums/for...=1685987788624
That means there are many missing data in those latter two variables (as not all the contracts span the duration of those two periods, for example a job contract contained in the database might have started in 1990 and ended in 1995, thus in durA and durB it would have only missing data, this would be scenario 1). Some contracts might have only values for durA (eg. a contract that begun in 2006 and ended in 2009) or for durB (eg. a contract that lasted from 2011 to 2013; this would be scenario 2). And finally (scenario 3), some contracts might have data for both durA and durB (a contract that began in 2000 and lasted until 2016 would have the maximum value for both durA and durB).
As far as I know, I can't perform a t-test on those two variables, as the data are only partially paired (those who fit into scenario 3). Many contracts only have values for either durA or durB. I've read a few papers on techniques to make comparisons involving weighted t-tests with partially paired data (eg. https://www.tqmp.org/RegularArticles.../p055/p055.pdf, https://digitalcommons.wayne.edu/cgi...&context=jmasm, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9041716/, https://citeseerx.ist.psu.edu/viewdo...=rep1&type=pdf), but there is no explanation on how to implement them on Stata (one of the papers mentions a package for R, though).
Would Welch's t-test work in this case? Or is there a way to perform the analysis using the main duration variable (the one that includes the duration of all contracts) but comparing the duration of only those two periods (that is, the values included between on one hand 2005-2010 and 2010-2015 on the other).
Thank you so much for your help.
Comment