The relationship between television (TV) viewing and overweight in children is likely to be influenced by the quality of measures used. We systematically reviewed studies related to overweight in children that had included a measure of TV exposure. Literature searches were conducted in MedLINE, PsychInfo, SportDiscus and ISIWeb of Science. Methods of assessing TV viewing were evaluated, including the type of measure, the administration procedure and reported psychometric properties. The majority of studies assessed TV viewing using self-report surveys and very few studies used direct observation. The validity or reliability of measures was often not examined. The majority of tools that were evaluated were compared with another self-report measure or an objective measure of physical activity. TV viewing measures should be selected that are specific to the research question, the study design, as well as methodological feasibility. However, it is recommended that measures of TV viewing be used only if they have psychometric data to support their validity and reliability. Selecting measures that are valid and reliable enables us to examine with greater accuracy the influence of TV viewing on childhood overweight, as well as the efficacy of interventions designed to reduce TV viewing in children.
It is estimated that children spend 1.8–2.8 h watching television (TV) per day (1). This has important implications for health and may be partly responsible for the rising prevalence of childhood overweight. Research examining the impact of TV viewing on childhood overweight became more popular in 1985 after Dietz and Gortmaker (2) found a positive association between TV watching and obesity among children. Since then, four mechanisms have been proposed in attempt to explain how TV impacts weight gain (i) TV displaces time that would otherwise be used for physical activity (3); (ii) TV viewing promotes between meal snacking (2) and therefore a greater total daily caloric intake; (iii) TV programme content exerts a negative influence on children’s food choices and attitudes towards health lifestyles through priming and/or cultivation (4) and (iv) TV watching decreases one’s metabolic rate (MET) (5,6). The relative contribution of each of these mechanisms to obesity is unknown.
The degree to which TV influences childhood overweight is controversial. Associations between overweight and duration of TV viewing are usually identified in cross-sectional studies, although these associations are often weak (7). It is possible that a lack of strength in association is due to inadequacies in study designs. It could also be argued that problems of TV measurement or variation in how exposure to TV is quantified contributes to discrepant findings and to the weak associations often reported in these studies (5,7–9).
We suggest that the methods for assessing TV have not kept pace with the research interest in this topic. The measurement of physical activity in children has experienced an increase in the use of objective monitoring (10), yet a concurrent increase in objective monitoring as indicator of inactivity or TV watching has not occurred. Objective measures of TV watching have been attempted in some small studies (11–14), yet variation in inter-rater reliability and difficulty coding participant behaviour contribute to error in this method as well (13). Self-report measures are commonly used, but research regarding their validity and reliability is limited, and it remains unclear whether one measure is superior to others.
To our knowledge, the psychometrics of the measures used to assess exposure to TV among children and adolescents have not been evaluated. It is therefore difficult for researchers to choose the most appropriate tool to measure TV watching in this population. The current aims of this study were to evaluate the methods used to measure children’s exposure to TV in studies focused on overweight (or related behaviours); and to summarize these methods’ measurement properties. To accomplish these aims, we conducted a systematic review of methods used to measure TV watching among children and adolescents in studies of weight, nutrition, physical activity, or physical inactivity. Such a review is intended to provide a resource for researchers to help determine appropriate measurement tools. Additionally, it is hoped that it will focus the field’s attention on improvements needed in measurement methods.
Search procedures and inclusion criteria
Relevant literature was obtained from the following computerized databases: MedLINE (PubMed), PsychInfo, SportDiscus and ISIWeb of Science (excluding arts and humanities). Keywords were: TV or television; and one weight or activity measure keyword (body mass index, obesity, overweight, nutrition, physical activity, sedentary activity, or physical inactivity).
Manuscripts published between January 1985 and May 2005 were included in the review, following the time when Dietz and Gortmaker (2) published their key article highlighting the importance of TV in the aetiology of childhood overweight in 1985. To capture only children and adolescents, the age of study participants was specified as being 18 years or less. Only peer-reviewed English-language articles were included. Reviews, position statements, case reports, abstracts and editorials were excluded as they lacked information about methods of measuring TV watching. We excluded cross-sectional or cohort studies with less than 100 participants and intervention studies with less than 50 participants. We did not impose a size restriction on studies which used objective measures, as these studies were generally smaller. Articles which met these criteria, but that lacked sufficient information describing the assessment of TV were also excluded from the review.
Other inclusion criteria related to the type of TV measure used in the study. Specifically, the measure was required to include frequency or duration of TV watching, and could not be manipulated by study investigators. TV had to be the main exposure or outcome of interest, which was operationalized by requiring that at least one relationship between TV and the other variables of interest be reported in the text or tables of the article.
The systematic review was conducted in two phases. In the first phase of the review, two independent reviewers evaluated each abstract based on the inclusion criteria. Abstracts were excluded from further analysis if both reviewers agreed that they did meet study criteria. In the second phase of the review, full articles of included abstracts were obtained and evaluated by both reviewers. For both phases of the review, inter-rater reliability was assessed using per cent agreement and unweighted kappa statistics. Disagreements about inclusion/exclusion were resolved by consensus. Validity and reliability estimates were obtained when available (Table 1). As a guide, we followed the ratings suggested by Landis and Koch (15) for agreement: 0–0.19 = poor, 0.20–0.39 = fair, 0.40–0.59 = moderate, 0.60–0.79 = substantial and 0.80–1.0 = almost perfect.
Table 1. Measures of television (TV) watching that were assessed for reliability (test–retest) and/or validity (criterion or construct)
|Tools exclusively designed to assess TV|
|Direct estimate of hours per week of TV||(14)|| || ||10-d TV viewing diary||Total cohort (n = 330) r = 0.60|
|Shannon 7-d TV grid||(25)||2 weeks||Subsample (n = 40) r = 0.68|| || |
|Robinson school-based intervention self-report instrument||(38)||Same day||Subsample (n = 80) r = 0.94 (range 0.79–0.93 for individual items) (P value not reported)||two-item self-report physical activity questionnaire||Weak negative association between physical activity and TV exposure (r = −0.04)|
|Sports, Play and Active Recreation For Kids self-reported TV viewing||(17)||4 d||Pilot sample of 69 students: r = 0.58 (P value not reported)|| || |
|11-item TV Video measure||(33)|| || ||24-h physical activity recall||Results from subsample of 53 students: r = 0.54|
|Global Weekly Estimate of TV viewing||(13)|| || ||Direct observation (video recorder)||Parental report more accurate than child self-report. Statistics not reported. Our calculations from data presented: mother r = 0.15; child r = 0.39|
|Summed Weekly Estimate of TV viewing||(13)|| || ||Direct observation (video recorder)||Parental report more accurate than child self-report. Statistics not reported. Our calculations from data presented: mother r = 0.34; child r = 0.07|
|San Diego Study of Child Activity and Nutrition interviewer-administered survey||(24)||2–3 weeks||Pilot study r = 0.80 || || |
|Sedentary measure for Project Eating Among Teens||(40)||2 weeks||Subsample (n = 167) weekday r = 0.80; weekend r = 0.69 || || |
|The Eating and Activity Questionnaire Trial (Project EAST): brief questionnaire to assess TV viewing and computer use||(39)||7 d ||Weekday summer r = 0.25; weekend day summer r = 0.27; and weekend day school r = 0.44||7-d TV diary/log||Average week r = 0.47; and average weekend day r = 0.37|
|Physical activity surveys with a TV assessment component|
|Daily activity chart||(14)|| || ||10-d TV viewing diary||Total cohort (n = 330) r = 0.48|
|Self-assessed physical activity checklist (SAPAC)||(45)||5 d ||TV/video: r = 0.20 boys, r = 0.38 girls. Computer: r = 0.40 boys, r = 0.35 girls. Total sedentary: r = 0.36 boys, r = 0.34 girls||Activity monitor (Caltrac), heart rate monitor||Activity monitor: self-report r = 0.30; interviewer-administered r = 0.33; Heart rate monitors: self-report r = 0.57; interviewer-administered r = 0.51|
|Modified SAPAC (7-d recall format)||(61)||7 d||Pilot sample of 100 students: r = 0.70 for seven sedentary items|| || |
|New moves obesity prevention physical education programme||(48)||1 month||Pilot sample of 56 adolescent girls: r = 0.80 || || |
|Girls health Enrichment Multi-site Studies Activity Questionnaire||(53)||2 weeks||Total cohort (n = 210). Previous day estimate of TV exposure r = 0.13; estimate of usual TV exposure r = 0.31||3-d activity monitor (Actigraph)||Total cohort (n = 210). Negative association between monitor counts and usual estimate of TV exposure r = −0.19|
|Previous day physical activity recall||(62)||1 h||r = 0.98||Pedometer, physical activity monitor (Caltrac), heart rate monitor||Validation for all activities in previous 24 h. Pedometer r = 0.88, activity monitor r = 0.77; heart rate monitor r = 0.32|
|Cebu Longitudinal Health and Nutrition Survey physical activity questionnaire||(54)|| || ||24-h activity monitor (Caltrac)||Negative association between monitor counts and self-reported TV: males r = −0.14; females r = −0.15|
|Robinson assessment of after-school activities||(60)||24 months||No-intervention control group (n = 279) r = 0.37||Three-item self-report physical activity questionnaire||Weak negative association between physical activity and TV exposure (r = −0.09)|
|Assessment of TV watching in national surveys|
|1999 Youth Risk Behavior Questionnaire||(39)||7 d||Single-item for weekday viewing r = 0.34||7-d TV diary/log||Single-item for weekday viewing r = 0.5|
|Icelandic survey||(71)|| || ||Physical activity score: two self-report items (hours/weeks)||Negative association between physical activity and TV exposure r = −0.06|
|Habit books with index cards||(77)|| || ||Activity monitor (Tritrac) for 4 d||Results from subsample of 41 children: r = 0.63|
|10-d TV viewing diary||(14)||1 month||Total cohort (n = 334) r = 0.72||Direct observation (video recorder)||Subsample (n= 99) r = 0.86|
We identified 845 articles using our search criteria. Based on the information contained within article abstracts, 504 articles were excluded for not meeting inclusion criteria. Inter-rater reliability for this phase was substantial (87% agreement; kappa 0.71). A further 234 articles were excluded from phase two of the search, based on a review of the full manuscripts. This resulted in 98 manuscripts being retained for evaluation. Inter-rater reliability for phase two was almost perfect (92% agreement; kappa 0.85). We divided manuscripts according to the method in which data were recorded: self-reported surveys; self-reported diaries; or direct observations of TV. Self-reported surveys were then divided into those where the method was (i) designed exclusively to assess TV; (ii) combined physical activity and sedentary behaviours and (iii) single-item TV assessment.
Table 2 shows the self-report methods used to assess TV watching in children and adolescents. A total of 88 studies were found using our search criteria. Some studies collected information on TV watching using more than one type of self-report method and were therefore entered more than once (i.e. they may have obtained both child and parental reports). Consequently, the number and per cent of articles that are reported to have used individual methods in Table 2 do not necessarily equal 88% or 100% respectively. The majority (78%) of surveys measured TV watching in children (aged 3–12 years), although many were also conducted in adolescents (55%). Just 5% of articles assessed TV exposure in infants (aged 0–3 years). Eighty per cent of studies asked the child to respond to questions and 29% used parental report. In general, studies assessing TV in younger children were more likely to rely on parental reports. We found 11 articles that had assessed both parental and child self-report of TV exposure (2,16–25). Four of these were completed as parent–child pairs together (16,18–20), and two asked parents to complete surveys for the youngest children in the cohort (2,21). The remaining five articles assessed parental and child self-reports of TV exposure separately (17,22–25). For those that tested the relationship between parent and child report, correlations were generally high (range r = 0.31–0.92), although parents consistency tended to underestimate the amount of time their child watched TV compared with the child self-reports (17,24).
Table 2. Methodologies employed for single-item and multiple-item surveys of television (TV) watching
|Child’s age (years)|
| Infant (0–2.9)||(32,47,66,84)|| 4 |
| Child (3–12.9)||(2,5,16–19,21–25,26–34,36,37,42,41–45,47,49–53,55–59−61,63,65–70,72–76,84–100)||69 |
| Adolescent (13–17.9)||(2,5,16,18–21,27–31,33–35,38–40,42,46,48,50–52,54,60,61,63,64,71–75,86,88,90,92,99–108)||48 |
|Self-report provided by|
| Child/adolescent||(2,16–20,22–25,30,34,38–41,53,54,63,71–75,86,88,96,97,99–102,107,108)||70 |
| Parent (primary caregiver)||(2,5,16–25,32,55,56,59,65,66,69,70,76,84,87,89,91,93–95,98)||29 |
|Mode of administration|
| Self||(16,17,20–23,25–34,36–49,50–53,55–59,60,61,63–66,68,70,71,84,86–88,90–95,98–109)||71 |
| Interviewer||(5,18,19,24,35,54,67,69,72–76,89,97)||15 |
| Usual duration||(2,17,21,22–34,38–42,48,52–55,58–61,65–71,85–88,90–98,100,101,103,105–109)||58 |
| Specific time period||(2,5,18–20,22,23,25,32,35–37,43–47,50,53,56,57,63,64,72–76,84)||29 |
| Not specified||(16,49,51,99,102)|| 5 |
| Day||(2,16,17,25,30,35,43–45,52,53,56,64,71–76,85,86,89,90,93,99,100,104,105)||28 |
| Whole week||(20,49,54,58,59,61,67,69,70)|| 9 |
| Weekend days separately||(5,18,19,22,24,26–29,32,34,37,38,40,42,48,50,55,57,61,65,66,86–88,91,94–96,98,107,108)||32 |
| Weekdays (school days) only||(21,39,41,84,97,101,103,106)|| 8 |
| Around school times||(22,23,36,45–47,51,60,63,64,87,96,102,104,109)||15 |
| Closed-ended||(2,21,27,29–31,33,38–40,43,44,48,53,55,60,61,71–75,92,98–103,106,108,109)||32 |
| Open-ended||(5,16,20,22–26,28,32,34–37,41,42,45–47,49–52,54,56–59,61,63–65, 67–70,76,84–91, 93–97,104,105,107)||53 |
|Unit of time for response|
| Minutes||(35,43–47,53,56,57,61,65,84,87,93,96,108)||17 |
| Hours||(2,5,17–34,36–42,48–52,54,55,58–60,63–66,68,69,72–76,85,86,88–92,95,101,102, 98–100,103–107,109)||69 |
| Duration from TV programme schedule||(67,68,94,97)|| 4 |
|Video games included in TV measure?|
| Yes||(16,20,22,23,36,37,46,47,56,57,59,61,64,90,104,109)||16 |
| No||(2,5,18,19,21,24–35,38–45,48–55,57,58,60,61,63,65–75,76,84–89,91–93,95–103,105–108)||72 |
|Surveys that exclusively measure TV watching||(13,14,22–25,27–41)||21 |
Most of the surveys were self-administered (81%), while only 17% were interviewer-administered. (Some used both methods.) Most (66%) studies used a recall frame of a ‘typical’ day, week, or month, with 33% relying on a specific recall period (e.g. yesterday, last week, etc.). Thirty-two per cent of the surveys obtained TV recall exposure for a single day, but the most popular method asked participants to recall weekdays and weekend days separately (36%). Most surveys (60%) included items that were in open-ended format, such that participants were required to estimate the time in minutes or hours that they watched TV during each recall interval. Other surveys (36%) used a quantitative response format in which participants recorded their TV viewing on an ordinal or ratio scale that was anchored by time (e.g. 0–1 h d−1, 1–2 h d−1, 2–3 h d−1, >3 h d−1) (26). The majority (78%) of studies asked respondents to report the amount of time watching TV in hours. Nineteen per cent of studies requested estimations in minutes and 5% obtained an estimate of TV exposure by asking the respondent to check which TV programmes they watched from a TV schedule. Although 18% of studies grouped TV and video game exposure together, most (82%) asked about TV viewing and video game playing separately.
Surveys that exclusively measure television watching
Approximately 25% of studies assessed only TV watching (13,14,22–25,27–41, see Table 2). Many were adapted from existing tools (24,27–30,35). The most popular method of recall was to separate weekdays from weekend days (13,24,25,27–30,32,37,39). Average weekly time spent watching TV was then calculated by weighting the sum of the weekend and weekdays. Alternative recall methods were performed by Janz and Mahoney (35), who asked participants to recall the previous day and evening; and Shannon et al. (25), who supplied a 7-d grid with 30-min blocks in which children were asked to recall TV watching.
Table 1 shows the psychometric properties of measures in which some form of evaluation was performed. Of these, we identified nine articles that measured only TV viewing (13,14,17,24,25,33,38–40). Test–retest reliability was assessed in six of these, with a range in correlation of r = 0.25–0.8 (retest period ranging from same day to 3 weeks). Criterion validity was examined in only one study (13), in which two tools were psychometrically assessed using direct observations with video recorders (range r = 0.07–0.39). Comparisons with other objective tools were not performed, but four articles correlated TV viewing with self-reported activity or inactivity (14,33,38,39).
Physical activity surveys with a television assessment component
Many studies incorporated the measurement of sedentary behaviours within a scale of more physically intense activities (14,21,26,42–60). Tools that were designed to assess physical activity and sedentary behaviours which were tested psychometrically included the Previous Day Physical Activity Recall (PDPAR) (43,44); the Self-Assessed Physical Activity Checklist (SAPAC) (45–47,61); the New Moves questionnaire (48); the Cebu Longitudinal Health and Nutrition Survey (CLHNS) physical activity questionnaire (54); the Robinson assessment of after-school activities (60); the Daily activity diary (14) and the Girls Health Enrichment Multi-site Studies Activity Questionnaire (GEMS AQ) (53) (Table 1). Test–retest reliability was performed for the PDPAR (r = 0.98; 1-h interval between tests) (62), SAPAC (range r = 0.34–0.7; 5–7 d between tests) (45,61), the New Moves questionnaire (r = 0.8; 1 month between tests) (48) and the GEMS AQ (r for previous day recall = 0.13, r for estimate of usual exposure = 0.31 with 2 weeks between tests) (53). Studies that evaluated construct validity by correlating activity time with sedentary behaviour time included the PDPAR (r = 0.77), SAPAC (r = 0.3 for self-report and r = 0.33 for interviewer-administered), the GEMS AQ (r = −0.19) and the CLHNS (r =−0.15). Construct validity was also assessed using heart rate monitoring for the SAPAC (r = 0.57 for self-report and r = 0.51 for interviewer-administered) and the PDPAR (r = 0.63). The Robinson Assessment of After-school Activities Survey was compared with self-reported physical activity using a three-item physical activity questionnaire, and reported a poor correlation of r = −0.09. Anderson et al. (14) reported a moderate correlation (r = 0.48) between TV viewing from a daily activity diary and TV viewing from a 10-d TV viewing diary (described in section Self-reported diaries).
Assessment of television watching using a single item
In a large number of the articles that were identified using our search criteria, TV assessment was a minor part of a multi-component survey; often as just a single item. Many nationally representative studies used this method, including the National Longitudinal Study of Adolescent Health (18–20); the Youth Risk Behaviour Survey (YRBS) (63,64); the National Longitudinal Survey of Youth (5,65,66); the National Heart, Lung, and Blood Institute (NHLBI) growth and healthy study (67,68); the National Longitudinal Survey of Children and Youth (69,70); the Icelandic Survey (71); the National Health and Nutrition Examination Survey III (14,72–76); and the National Health Examination Surveys II and III (2). All of these obtained an estimate of the number of hours that TV was watched from a single survey item, except for the NHLBI growth and health study, in which estimates were gathered from participant reports of watching specific programmes from a weekly TV schedule. We were only able to identify two studies that evaluated any of these for reliability or validity. Schmitz et al. (39) compared self-reported weekday viewing from the YRBS with self-reported weekly viewing diaries and reported a moderate correlation (r = 0.5). Reliability for the same item completed 1 week apart was fair (test–retest r = 0.34). Vilhjalmsson and Thorlindsson (71) reported poor negative correlations with their TV exposure item and a self-reported physical activity questionnaire (r = −0.06).
Table 3 shows eight studies that were identified using our search criteria which assessed TV watching prospectively by use of a viewing or activity diary (9,14,25,39,77–80). Parental report (often with the help of the child) was obtained in five of the eight articles (9,14,25,77,78). Intervals in which participants were asked to report activities ranged from 1-min to 16-min blocks and they all differed with respect to the number of days that records were kept (ranging from 24 h to 2 weeks).
Table 3. Self-reported diaries of TV watching, in order of child’s age
|Vandewater et al. (2004) (9)|| 0–12||2831||Parent||24 h||Open (minute)|
|Anderson et al. (1985) (14)|| 5|| 330||Parent||10 d||15-min intervals (6 am−2 am)|
|Epstein et al. (2004) (77)|| 8–12|| 63||Parent and child|| 4 d||10-min intervals|
|Epstein et al. (1995) (78)|| 8–12|| 55||Parent and child|| 2 weeks||Open (minute)|
|Myrtek et al. (1996) (80)||10–13|| 50||Child||23 h||10–20-min intervals†|
|Schmitz et al. (2004) (39)||11–14|| 245||Child|| 1 week||30-min intervals|
|Shannon et al. (1991) (25)||11–12|| 773||Parent|| 1 week||Not stated|
|Katzmarzyk et al. (1998) (79)||12–14|| 183||Child|| 3 d||15-min intervals|
Two diaries were evaluated for psychometric properties (14,77). Anderson et al. (14) recruited 334 families, and asked parents to complete two 10-d viewing diaries (1 month apart). Parents were asked to report whether the TV was switched on, what channel was playing, the title of the programme, who was in the room and why the child stopped watching. A separate page was also provided to assess away from home TV viewing for the child. The accuracy of the 2nd 10-d viewing diary was assessed using video recorders that were placed in a subsample of 99 participant homes. Inter-rater reliability was high for the presence of the child in the TV room (r = 0.98) and for whether the child was paying attention to the TV (r = 90). TV exposure calculated from the diaries was found to correlate with direct observations (r = 0.86) and test–retest reliability was substantial (r = 0.72).
Epstein et al. (77) developed a 4-d diary (i.e. habit book) in which both the parent and child were asked to report any activity that the child did for at least 10 min. Behaviours were not specific to TV watching and included both physical activity and sedentary behaviours. A subsample of 41 children were fitted with activity monitors to assess construct validity. Energy expenditures, as calculated in multiples of the MET for total activities from the self-report and activity monitor, were substantially correlated (r = 0.6).
Direct observation of television viewing
We found only five studies (11–14,81) that measured TV viewing directly (Table 4). Of these, three relied on direct observation (one or two persons) (11,12,81) and two relied on videotaping (13,14). Each study recognized the variability of TV watching and thus for all studies the monitoring time period covered multiple days. The definition of TV watching varied across studies and only one study also collected video gaming as a separate measure (13). Four of the five studies assessed test–retest reliability, which was uniformly very high (11–14). One study reported participant recruitment rates for those asked to have video recorders in the home compared with those not asked. Recruitment was 15% lower for participants who were asked to allow in-home observations of TV exposure, perhaps indicative of the intrusive nature of direct observations.
Table 4. Objective measures of television (TV) watching, in publication date order
|Child’s age||5 years||4 years||3–4 years (1994) and 5–6 years (1996)||Third to fourth grade students|
|Sample size||334 (99 had video installed)||351||191 (1994) and 138 (1996)||10|
|Mode of observation||Video||Direct||Direct||Video|
|Number of observers||N/A||One||Two alternated every 2-h||N/A|
|Frequency||Only when the TV was on over 10 d||Two home and two pre-school/recess observation periods in a 2-week period||6–12 h d−1, up to 4 d year−1 over 3 years||Over 10 d|
|Sampling unit||55 min||1-h per observation period in the home||1 min||5 min|
|Definition of TV watching||(i) in room and TV on; (ii) in room, TV on and child is visually oriented to the TV||Child attending to the TV at the end of the observation interval||Minutes TV was on, child is in room and child is attending to the TV||In TV room looking at TV (other definitions also used)|
|Video game measured separately||No||No||No||Yes|
|Inter-observer or rater reliability||(i) r = 0.98; (ii) r = 0.90||No||96% agreement||kappa = 0.92|
The contribution of TV watching towards overweight in children has gained popularity over the past decade. We found 98 articles that had reported exposure to TV as a primary outcome, in addition to a selection of review articles describing the relationship between TV and overweight. However, few of these used a tool to measure TV exposure that had been psychometrically tested. We identified 14 measures (14% of total 98 manuscripts) that had been examined for test–retest reliability and 15(15% of total) that been assessed for some form of validity. Of these, seven (7% of total) were assessed for both reliability and validity (14,38,39,45,53,60,62).
Test–retest reliability ranged from r = 0.13 (previous day self-report in the GEMS AQ, 2 weeks apart) to r = 0.98 (previous day self-report in the PDPAR, 1 h apart). Validation of measures using a criterion method (e.g. direct observation) was only performed for three measures (i) Global Weekly Estimate of TV viewing (13); (ii) Summed Weekly Estimate of TV viewing (13) and (iii) the Anderson et al. 10-d viewing diary (14). Statistical correlations of self-report and direct observations were only reported for the 10-d viewing diary (r = 0.86 completed by child and parent together). Smaller participant numbers are expected in the validation of TV measures because of the intense level of video interpretation and coding that is required. However, small sample sizes preclude a statistical analysis of validity. We calculated correlations between self-reported TV and direct observations from the raw data for the other tools and found a range in validity from r = 0.07 (child self-report of the Summed Weekly Estimate) to r = 0.39 (parental report of the Global Weekly Estimate of TV viewing). Thus, based on text provided by the authors and our correlations (Table 1), the Summed Weekly Estimate appears to be most accurate if completed by the parent and the Global Weekly Estimate seems to be more precise with child self-report methods. These conjectures are based on estimates from a small sample of participants and further evaluation of both tools is recommended.
A number of studies assessed construct validity of TV watching against an objective (53,54,62,77,82) or subjective (14,33,39,71) physical activity measure. Rather than assessing the accuracy of the measure itself, this method relies on the degree to which the data correspond with behaviours that are related to TV. Such comparisons may be insufficient in terms of assessing the true validity of the measure. For example, researchers might hypothesize that high exposure to TV relates to less time spent performing physical activity. This was supported in studies that focused on TV items, whereby negative associations were observed between objective measures of physical activity and TV watching (53,54). A small negative correlation between physical activity and TV has been previously reported in a meta-analysis of media use, body fatness and physical activity in children (7) with a mean effect size of −0.10 (95% CI = −0.08 to −0.11). It is possible that such findings are dependent on the intensity of physical activity, as it has been recently theorized that participation in intense physical activity is compensated by an increased amount of sedentary behaviour (83). Given this, it would seem more appropriate to correlate self-reported TV watching with a measure of inactivity rather than activity. Furthermore, weak correlations between TV viewing and physical activity may result from the capacity to engage in both activities, as it is likely that TV viewing occurs mainly in the evening when physical activity is usually restricted in children (110).
Six articles reported correlations with other self-reported measures (14,33,38,39,60,71). Evaluation of one self-report method using another self-report method may result in an amplification of errors. For example, some studies compared a TV recall with a diary or log completed prospectively. While a diary may be less subject to memory bias, it suffers from other measurement errors that are common to those observed in all recall methods (e.g. social desirability) and may be influenced more greatly by compliance. Few studies employed a method of objective monitoring of TV watching. We identified only four studies that relied on either direct observation or videotaping (11–14,81). Both methodologies are invasive and not practical in large-scale research studies. Alternative objective measures are needed.
When choosing an appropriate method to assess TV exposure, the researcher should consider the level of precision that is required, the sample, as well as feasibility issues. For example, studies aiming to accurately assess TV exposure prospectively in young children may request parental report, or parental report with the assistance of the child. Also, prospective assessment allows the collection of more information, such as the type of TV programme and the reason that the child stopped viewing. However, this method is more time- and cost-intensive than retrospective reporting.
Research that is focused on TV watching in children should consider using a tool that was developed to measure TV exclusively. The measure with the greatest reliability was the ‘Robinson school based intervention self-report instrument’ (38), with an almost perfect correlation of r = 0.94 in a sample of 80 children. However, tests were administered on the same day and this may serve to inflate the reliability estimate. The measure with the highest validity correlation was the ‘Direct estimate of hours per week of TV’, developed by Anderson et al. (14). In a sample of 330 5-year-old children, a substantial correlation of r = 0.6 was reported between it and a 10-d viewing diary. Regrettably, this was not compared with a criterion method of direct observation.
Studies that are aimed at examining behaviours that relate to healthy weight in children could use tools which assess both physical activity and sedentary behaviours, with a TV component. This type of assessment was used by the majority of articles that we identified. Few had performed any type of psychometric assessment. In those that did, none assessed validity with a criterion measure. The measure that demonstrated the greatest reliability was the PDPAR (r = 0.98) (62). Again, tools that are tested twice within the same day are likely to result in higher correlations than those with a longer interval between administrations; thus, findings are more likely to represent consistency of viewing rather than stability across time and benefit by reducing the possibility of measuring real behavioural change. Another example of a tool with good reliability was the ‘New Moves obesity prevention physical education program’ (48), which was administered 1 month apart in a sample of 56 girls (r = 0.8). The PDPAR was also the tool with the best validity (62). The correlation between the whole survey (including physical activity questions) and pedometer counts was r = 0.88. As this was not specific to TV watching, we cannot infer the accuracy of the TV items specifically.
Many studies obtained TV watching using single items within a survey. Single-item use for the assessment of TV lacks content validity, is subject to measurement error related to memory and social desirability bias and only results in a crude estimate of TV viewing. It also cannot assess both frequency and duration of TV watching. Hence, it is not recommended for researchers whose primary outcome is exposure to TV or change in viewing following an intervention. It may, however, be appropriate for surveillance studies. We found just two surveys in which the TV measure had been by psychometrically evaluated (39,71), indicating a critical gap in the literature.
Activity or viewing diaries enable the collection of more detailed information such as the title of the TV programme or specific time periods that the TV is turned on. Also they are not affected by memory. Acquisition of detailed information using this type of method could also be used as a means to quantify participant compliance. The 10-d viewing diary developed by Anderson et al. (14) has been evaluated for both reliability and validity (with direct observations) using large samples (n = 330 and n = 105 respectively) with substantial to almost perfect correlations observed for both (r = 0.72 and r = 0.86 respectively). The greatest limitation with this method, however, is that viewing/activity diaries are intrusive and may induce behavioural reactivity.
Direct observations (either by video or by researcher) are considered a gold standard method for measuring TV exposure. For the majority of studies, this methodology would be considered infeasible, as it is expensive and requires a high level of demand from the researcher and the participant. Direct observation from researchers is also likely to influence behaviour; resulting in exposure estimates that are affected by social desirability. Even video-recorded observations may influence findings via participation bias (14), and multiple recordings are required in modern households, which often have TVs in more than one room. There is urgent need for the continued development of observational tools; in terms of both technology and feasibility. Ideal measures should not rely on self-report, and should have low participant and research burden.
Strengths, limitations and conclusions
Strict adherence to our search criteria (e.g. sample size restriction and our focus on obesity-related topics) meant that we undoubtedly missed some articles that assessed TV in children and adolescents. It is possible that key manuscripts published before or after our search dates were missed, or that our criteria to include only those articles which were focused on healthy weight behaviours in children precluded articles which were purely driven by methodology. Further, it is possible that a number of articles were excluded by our sample size requirements. Some articles cited previous manuscripts to supply information on the design and/or evaluation of measures. These were obtained for discussion, although they were not included in Table 2. In addition, we did not include articles that provided an insufficient description of the method used or a reference to find the methodology. Exclusion of such articles have precluded us from drawing conclusions based on these manuscripts; however, given the nature of this article, a degree of methodological information was essential to assess the merits and shortcomings of each measure. Many articles failed to fully describe all of the study procedures. We made no assumptions when this occurred, which ultimately resulted in a reduced specificity of this review. This review was systematically conducted by multiple reviewers. As we were able to locate a large number of studies using a thorough search procedure covering a period of almost 20 years, we are confident that we were able to obtain an adequate representation of the measures currently available. We did not examine whether researchers assessed the location of TVs in the home. While there is a lack of evidence in this area, it is likely that the number and location of TVs in family homes relates to the duration and frequency of viewing. Validation of tools using direct observation should consider viewing TVs in more than one room. Further, consideration should be made to whether the child is actually sedentary while they are watching TV, as there is some evidence that TV viewing is often accompanied by other tasks (111). To our knowledge, this is the first article which has attempted to evaluate methods that measure TV exposure in children specifically. Given the alarming rate of increase in childhood overweight and obesity, the development of accurate assessment tools to measure suspected determinants of excessive weight gain in young people, such as TV viewing, is essential.
Conflict of Interest Statement
No conflict of interest was declared.
Support for this research project was provided by an unrestricted gift from the Gatorade Company for the Get Kids in Action Partnership with the University of North Carolina – Chapel Hill.