Hi,
I’m working with MTUS data and I would like to generate population-level estimates (i.e. the sum of hours spent on housework for the total adult population). I’ve been reading about the propwt weight, which is provided in the data, and it says that it deflates the original weight (when the original weight inflated to the population size). Does the MTUS include the necessary information to reconstruct the inflated weight? Is this in the OCOMBWT weight? In sum, what is the best way to generate population-level estimates using the MTUS weights?
Thank you!
Pilar
PROPWT is based on the original survey weight, OCOMBWT; you can read about how it is calculated in the description tab for PROPWT. You are correct that the original survey weight is initially deflated in order to calculate PROPWT based on a good-diary inflation factor. Therefore, PROPWT should have a mean of 1 and sum to the total number of good diary cases. OCOMBWT is the appropriate weight to use in order to generate national population estimates, though important comparability issues should be considered because it inflates samples differently based on the original survey.
Thank you for your answer, Grace! I’m perhaps missing something but I still seem unable to generate population-level estimates with PROPWT. The values for this variable range from 0 to 12 only, and with about 10000 observations per year in the US even if each observation “represented” 12 people it would only add up to ~100,000. Thus, it seems that an inflation factor is still missing to scale up to the population level (i.e., calculating a weighted N that equals the country total population). Let me know if you have other details. Thank you!
You are correct that PROPWT does not generate population-level estimates; it inflates to the number of valid diary entries. My previous response will be edited to ensure that accurate information is given to future readers of this post. OCOMBWT is the appropriate weight to use in order to get national population estimates. According to the description tab for OCOMBWT, “If the original weight is inflated, OCOMBWT is left inflated”. It is important to consider that there are comparability issues with OCOMBWT because it inflates samples differently based on the original survey and because the specific target population will vary by sample. Also, keep in mind that this weight may include “bad” diaries if the original weight did not exclude them.
As a sanity check, I created extracts using both ATUS and MTUS in 2018 to compare population estimates using WT06 in ATUS and OCOMBWT in MTUS; they provide matching population estimates. You can find more information on specific samples from the MTUS source documents page, and may be interested in the weighting-specific information in the MTUS Coding Procedures document from the IPUMS Time Use team’s colleagues at the Centre for Time Use Research.
1 Like
Hi Grace,
Thanks so much for your the informative answer re MTUS weights. I have a similar question/issue with AHTUS weights. I want to generate population-level estimates and after reading the files it appears that INFLTWT is the right weight to do this. When I use this variable, however, I encounter two problems: 1) the values do not appear to add-up to population-level estimates (I do not get the correct population size, for instance), and 2) the values do not appear consistent across years (I obtain values way below the population size before 2003 and values way above the population size starting in 2003). Do you have any suggestions on how to use the available weights to obtain population-level estimates for the full time-series (or at least for 1965-2000)? Thank you!
Pilar
Weights in IPUMS Time Use are a bit different than other datasets available through IPUMS in that sample weights indicate the number of person-days the respondent represents (not the number of people). This may be the source of your discrepancy. You should also ensure that you are comparing the same population over time as the sample characteristics for AHTUS data vary over time (particularly before 2003). More information on weights in AHTUS are available beginning on page 9 of the AHTUS User Guide. While INFLTWT retains the ATUS inflation factor for 2003-forward samples and is designed to reflect the CPS distribution for earlier years of data, you may be interested in XTIMEWT for an analysis that spans this broad range of years.
1 Like