# Seemingly missing values for time variables and weights using Stata

Hello,

I have two questions regarding the use of the ATUS time use data, that I’ll be thankful for your response to:

1. I couldn’t find a guide regarding the use of weights (the ordinary person weights) in Stata. Are there some instructions for that?
2. I encountered why I think is an odd phenomenon. I looked at time variables for time spent with children and limited the analysis only to households in which there is a child (or more than one). However, there is still a large share of the sample - about 22%, who report 0 time spent with children although they live in a household in which there is a child. This seems rather unlikely, so I’d like to ask whether you know something about a possible reason for that (that will allow me to determine whether I can drop these observations from my sample).

Thank you very much

Hi Assaf,

I hope I can help.

1. Weights are briefly summarized on the FAQ page. In general, you should use WT06 together with the replicate weights found in RWT06 for empirically derived standard errors. There is sample code for replicate weights on the CPS replicate weights page. You should use WT20 together with RWT20 when analyzing the 2020 sample. All of the weights are probability weights. In stata, you would run your regression with [pw=wt06].

2. Finding total time spent with children can be tricky. You’ll need to use AGEW to examine the ages of people the respondent engaged in activities with. Summing the number of occurrences of AGEW < 18 and tabulating them by HH_CHILD, you should find that among ATUS respondents in 2019-2021 with household children only about 5% did not engage in any activities with children under 18. Taking it a step further, respondents engaged in primary and secondary childcare sometimes omit that they were with a child at the time. When including households with ACTIVITY codes 30100-30399 (primary childcare for household children) and positive values for SCC_HH_LN (secondary childcare for household children), you should find that less than 5% of respondents in households with children report not being engaged in either primary or secondary childcare or any other activities with those under 18. This seems reasonable considering the respondent is a random member of the household.

Hi Ivan,

Thank you very much for your response. Just to make sure I understand:

1. I set the data using the code in the attached guide, and then use the [pw=wt06] command in my regression, or is it included in the command and I only need [pw=wt06]?

2. I created a variable for time spent with children using the “with whom” option. In that case, should it include both primary and secondary activities?

Thank you very much

1. The weights are not set in the command file. After setting the data, you need to include [pw=wt06] every time that you run a regression in which you want the results to be weighted and representative of how the population spends their time. Alternatively, you can run the command svyset [pw=wt06] so that stata remembers to weight all of your future estimates.

2. Yes, you will want to separately account for primary and secondary childcare since some respondents omit the with whom variable for these categories. Note that spending time with a child does not mean the individual is under 18 years older. Rather, it means they are the respondent’s child. In fact, the with whom option is not particularly helpful if you’re looking for all time spent with those under 18 since brothers, sisters, grandchildren, and other related people can all be under 18. I recommend instead creating a data extract with the variable AGEW to filter with whom by age rather than relationship.

Hi Ivan,

Thank you very much for your response. I wish to ask more about the last part regarding time with children variables. I am interested in the time parents spend with their children (under age 18). When it comes to specific activities seems fairly easy, as I can construct them using the activity option and specifying which activity I am interested in, and it seems to already differentiate between household and non-household as well as between children and adults. My bigger concern is about the total time spent with already - so far, I constructed a variable using the with whom option and setting an age limit. Is that correct, or should I also separately account for secondary activities?

Thank you very much

Hi Assaf,

Thank you for your patience. This is a difficult question, so we wanted to get it right. If you want to account for all time spent by respondents with children, you will definitely want to consider childcare provision instead of using only the “with whom” feature of the Time Use Variable (TUV) builder. While this feature allows you to select time spent by relationship and age (in the case of own household children), it will not capture all time respondents spent with children. This is because respondents often omit children from the list of people who they participated in the activity with if the activity name directly mentions children. This is particularly true in cases of secondary childcare; parents may view the question of who they are participating in the activity with as asking who is helping them with childcare and implicitly exclude children.

I can think of two different options for trying to capture all time spent with children: using the tool to create multiple TUVs that are mutually exclusive and summing them in your stats package, or requesting a hierarchical file and writing the code to build your own TUV. If you decide to keep using the tool, you will want to create three separate TUVs: a) All activities that respondents report spending with their child under 18, b) all activities that are directly related to the care of children under 18, but the respondent does not report being with a child under 18, and c) all instances of secondary child care where the respondent does not report being with a child under 18. Note that each of these methods will define “children” differently. The with whom feature is based on a specific relationship between the respondent and the other person(s) present during the activity. In contrast, primary activity codes for providing child care only specify with household children or non-household children (i.e., they are not contingent upon the relationship between the respondent and the child), and there are a number of different secondary child care measures to choose from. Additionally, you should look at the age restrictions for each measure. Once you’ve downloaded the data, sum the values for each of the three cases together. This will be the total time sample respondents spent with children.

To restrict your TUV more narrowly based age and relationship combinations, you could instead create a hierarchical extract with the variables ACTIVITY, DURATION, RELATEW, AGEW, as well as the secondary childcare measure(s) of interest. I also recommend taking a look at the ATUS training materials, which include exercises on creating TUVs from hierarchical files and another for thinking about childcare