Calibrating Stylised Time Estimates Using UK Diary Data

Abstract

This article proposes a method to improve the quality and potential of stylised time-use estimates with information derived from time diary data. Stylised time-use questions (direct questions about frequency of daily activities or "usual" hours spent on paid work and domestic work) are commonly incorporated into national surveys. However, stylised time-use data tend to contain both random and systematic errors (e.g., Jacobs 1998; Kan 2008; Kan and Pudney 2008). Moreover, they usually do not cover all types of daily activities. Even if they do, as in some UK 1980s datasets, they will add up more than 1,440 min per day (Gershuny and Robinson 1994). While the time diary collects accurate and detailed information about time-use on all major daily activities, keeping a diary is too time consuming and burdensome to be included in most surveys, especially panel surveys. To combine the strengths of a panel survey and diary data, we examine data from two UK datasets: (1) the Home On-line Study (HoL, 1999-2001), which collected both stylised and time-diary data from the same respondents, and (2) the British Household Panel Survey (BHPS 1991-2005), a major household panel survey in the UK. Results derived from HoL are used to calibrate time-use estimates for BHPS respondents. Unlike traditional data imputation techniques, which merge evidence across datasets using variables unrelated to the variables-of-interest, the potential of stylised time-use data in the BHPS are enhanced by matching them in a survey which collects both diary and the same stylised questions. We include a wide range of stylised variables (about paid work, leisure activities, core housework, and other domestic work activities) in the equations to estimate the time spent on all the main categories of daily activities. It is often difficult for respondents to recall and average their usual time spent on a particular activity in stylised questions. The inaccuracy of reporting could be reduced by incorporating a series of variables concerning the respondent's entire time budget on the right hand side of the equation. If one assume that the errors in the stylised reports of various different activities are at least partly uncorrelated?which is likely where, as in what follows, one has dif ferent sorts of stylised indicators for these various activities?then, since the various activities in a day in reality sum to a constant 1,440 min, stylised estimates of all the other