Psychometric validity of the parent’s outcome expectations for children’s television viewing (POETV) scale

The POETV was developed based on semi-structured interviews with mostly Hispanic parents of self-reported overweight or obese 5-8 year old children regarding the reasons why children watch TV and reasons why parents may allow or limit TV viewing (36). Multiple themes emerged. Parents anticipated both positive outcomes of allowing their child to watch TV (e.g. a safe activity for children while parents get their work done, and fun entertainment for their child), and negative outcomes (e.g. affecting the child’s health, influencing family relations and exposing the child to inappropriate content) [36]. Based on the interviews, three of the authors developed 34 items that encompassed outcome expectations that the parents reported: 17 that were perceived as positive and 17 that were perceived as negative. Instructions to the instrument read: “How much do you agree or disagree with the following statements. If I let my child watch TV…” Response options were a 5-point Likert Scale (1- Strongly Disagree, 2-Disagree, 3-Not Sure, 4-Agree, and 5-Strongly Agree). The English version was translated into Spanish by a staff member fluent in Spanish, and independently back-translated into English by a second staff member to ensure content equivalence between the English and Spanish versions. Differences in the original and back-translated versions were reviewed and consensus was reached for conceptual and cultural rather than linguistic equivalence. Ten Hispanic parents participated in cognitive interviews while completing the 34 item POETV, 5 who completed the questionnaire in English and 5 who completed it in Spanish. There were no resulting changes in the POETV instructions, items or response options.

A cross-sectional multi-ethnic sample of parents with children 6-12 year old (n = 311) completed the study survey available on the internet. Inclusion criteria were 1) parent or legal guardian of a 6-12 year old child; 2) parent lives with the 6-12 year old child at least 50% of the time; 3) resides in Harris County, Texas (location of Houston); and 4) able to read and write in English or Spanish. Exclusion criteria included parent of a 6-12 year old child with medical or behavioral problems that prevents him/her from participating in age appropriate activities. The survey was developed on Survey Monkey GOLD software ©, such that the URL of the survey was SSL/HTTPS encrypted for data security. The online survey was available in English and Spanish.

Parents were recruited via flyers or short presentations introducing the study with the survey URL link, at community centers, community businesses (gyms, apartment complexes, restaurants, museums), WIC offices, libraries, pediatric clinics, the Texas Medical Center, universities and community college bulletins, health and food fairs, and the Children’s Nutrition Research Center (CNRC), Baylor College of Medicine (BCM), and Texas Children’s Hospital websites. Active CNRC research volunteers were informed of the study by email or phone. Interested parents accessed the online survey and informed online consent was obtained at the start of the survey with an introductory letter, followed by a set of pre-screening questions to assess eligibility. Only those parents who provided consent by clicking “I agree to participate” and met eligibility criteria could access the online survey. The survey was available on the internet from April 2012 to August 2012. The survey was accessed by 595 people, 486 agreed to participate, of whom 91 did not qualify per the screening protocol. Of the 395 who qualified, nine participants were removed from analyses because they completed the survey more than once, leaving 386 participants. This analysis included 311 parents (80.6% of those that accessed the survey and qualified) who completed the POETV portion of the survey, 94.2% who were mothers and reported mostly male children (57.9%). Participants had the option to opt into a drawing for one of 15 $100 gift cards as a thank you for participating. The demographic descriptors can be found in Table 1. The study was reviewed and approved by the Baylor College of Medicine Institutional Review Board.

In addition to the new POETV instrument, parents completed a demographic survey, and reported on their child’s screen media use. TV viewing by the child was assessed by parent report of 5 items that were a modified version of a global weekly TV viewing estimate that had reasonable correspondence to videotaped observations of children’s TV viewing in the home [37]. Each item was asked separately for weekdays and weekends to assess: 1) Watching TV on TVs, computers or other devices; 2) Watching videos or DVD’s; 3) Playing videogames (such as Xbox, Wii, Playstation); 4) Playing hand-held videogames (such as a DS, iPod or Leapster); 5) Using a computer for something other than activities related to school or work (such as browsing the internet, computer games, or e-mail). Parents responded by entering the average number of hours and minutes per day for each screen use question (or a ‘zero’ if child did not use that screen technology). For this analysis, we combined watching TV and videos or DVD’s into a “TV viewing” variable. Playing videogames, playing hand-held videogames, and using a computer for non-academic activities were combined into “other screen media” variable. Three outliers were removed from the TV viewing analysis who reported TV viewing or other screen media use for greater than 20 hours per day, which was implausible. Test-retest reliability of the TV and other screen media use from another unrelated study were 0.68 for TV viewing on weekdays, 0.89 for TV viewing on weekends,0.57 for other screen media use on weekdays, and 0.93 for other screen media use on weekends (O’Connor 2014 unpublished), suggesting moderate to high reliabilities.

The percentages, means and standard deviations were calculated for all demographic variables and screen viewing behaviors. A priori, we elected to assess the factor structure for Positive and Negative POETV in two separate models. Our qualitative data suggested that parents identified both positive and negative outcomes for allowing their child to view TV [6], suggesting that they are not opposite ends of the same construct, but two separate constructs. Attempting to assess the factor structure of a combined model, may therefore be overly complex. Similarly, other studies have identified separate models for effective and ineffective vegetable parenting practices [38, 39]. One item from the Positive POETV was dropped prior to analysis being run based on face validity (“…he/she would learn to speak better English.”), because it was deemed to only apply to a sub-section of the sample. This left 16 items for the Positive POETV factor analysis.

We randomly split our sample in two prior to running the factor analysis. Exploratory factor analyses (EFA) on the first half of the sample were conducted using principal components with a varimax rotation on Positive and Negative POETV items using scree plots to determine the number and structure of factors. Scree plot analysis suggested a 2 factor solution for both positive and negative POETV models. EFA was conducted using SPSS (Version 22.0, Armonk, NY: IBM Corp). Items which did not load on any factors (factor loading < 0.4) or loaded on more than one factor with factor loadings > 0.4, [40] were deleted from the scale and the EFA was rerun with the reduced set of items. This process was iterated until an optimal result was obtained. To evaluate the EFA model fit, items were submitted to confirmatory factor analyses (CFA) on the second half of the sample using Mplus (version 7.11, Los Angeles) with the Mean-adjusted Weighted Least Square (WLSM), which is appropriate for the evaluation of ordered categorical data. Model fit for the CFA was tested using Hu and Bentler’s two-index strategy [41], which includes the comparative fit index (CFI), the Tucker-Lewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean squared residual (SRMR). According to Hu and Bentler [41], good model fit should meet one of the following two-index criteria: (1) TLI ≥ 0.96 & SRMR ≤ 0.09; (2) RMSEA ≤ 0.06 & SRMR ≤ 0.09; or (3) CFI ≥ 0.96 & SRMR ≤ 0.09.

Internal consistency reliabilities (Cronbach’s alpha and mean inter-item correlations) were calculated. The acceptable Cronbach’s alpha, and mean inter-item correlation are values greater than 0.7 [42] and 0.2 [43], respectively. Spearman partial correlations (due to non-normal distributions) were calculated among Positive and Negative POETV factors; and for children’s weekday and weekend TV viewing and all other screen media use. The partial correlations controlled for child’s age, ethnicity and parent education, since these variables have previously been associated with child TV viewing. The internal consistency reliabilities and Spearman partial correlations was conducted in SAS 9.3 (New York 2011).