Appropriate to Analyze Reading Intensity with ATUS?

The distribution for personal-interest reading for reading is terribly skewed. Scads of zeros and then a handful of MASSIVE values. It doesn’t seem responsible to report the average. It makes more sense to look at the share of folks reading at different intensity levels (0-5 minutes, 6-20 min., etc.).

Is it appropriate to use ATUS for such an analysis? Every article I’ve encountered reports the average. I fear this is because the weighting or something about the sample makes the sort of analysis I’d like to do a no-no.

Many thanks in advance for any guidance.

The reading for personal interest activity code (120312) includes:

-Being read to

-Browsing at the library

-Checking out library books

-Doing research

-Flipping/leafing through magazine

-Listening to books on tape/audio books

-Reading a book on a Kindle or other electronic book reader (2011+)

-Reading a magazine/book

-Reading scripture

-Reading the Bible

-Reading the newspaper

-Reading, unspecified

-Returning library books

It is specified that all of these activities must be done for personal interest.

I took a look at the distribution of time spent on reading for personal interest in all ATUS samples available from IPUMS ATUS (2003-2022). I see that one percent of respondents spent more than 4.4 hours, nine percent of respondents spent more than one hour, and 75 percent of respondents spent zero minutes on activities under reading for personal interest. The distribution does not lead me to believe that there is anything wrong with the data, or that we are reporting it inaccurately based on the time use recorded by respondents in their time diaries.

Time use surveys are known to include some amount of measurement error, as respondents do not perfectly recall or report their activities. IPUMS does not generally modify responses that seem unlikely based on their distribution across the sample. There are, however, respondents that are dropped from the sample because they provide implausible responses about their individual use of 24 hours. All time diaries that are included in IPUMS ATUS account for exactly 24 hours. This means that each respondent reports what they did during one day, which may or may not have been a representative day for them. For instance, a person who usually reads for personal interest for one hour a day could have been surveyed on a day that they read for much more or much less than one hour. But across the dataset, the distribution of the amount time spent on different activities should be representative of the population.

Weights must be used with all analyses of microdata samples. There are several features of the sampling and data collection process that make weights necessary. First, people and households with some characteristics are over-represented in the data, while others are underrepresented. Second, weekend days are oversampled - half of the respondents report activities that occurred on a weekend day while the other half report on activities that occurred on a weekday. In addition, response rates differ across demographic groups and days of the week. For these reasons, some ATUS records represent more cases than others and weights allow you to take this into account when analyzing the data. WT06 (a probability weight) should be used with most analyses with ATUS data, and RWT06 (a replicate weight) should be used in conjunction with WT06 to estimate standard errors.

Each individual researcher must make subjective decisions about use of weights, which parameters to estimate (mean, median, mode, distribution, etc.), how to restrict their analytical sample, and other choices which affect their final calculations. It is appropriate to estimate means and distributions of time spent on different activities using ATUS data, as long as weights are used appropriately and data is cleaned. There is no reason ATUS data should not be used to estimate the share of respondents whose responses fall within a specified range. If you choose to systematically exclude respondents (for example, by reporting the mean amount of time spent on reading among people who read at all in the day represented in the ATUS), you should subset your analysis appropriately (by using svy subpop in Stata, for example).

Many thanks for your detailed and timely response, Isabel! Such a help.