Measuring the Impact of Advertisement on Children Diet Choice in the Pediatric Population.

Background: The aim of this study was to evaluate the test–retest reliability of the paired selection method for measuring the impact of food advertising on the consumption preferences of children. Methods: In this study, 52 children aged 4 to 6 years participated in watching an advertisement style video showing four foodstuffs with low nutritional value. Two pollsters were introduced at two instances showing pairs of pictures of the advertised foodstuffs right next to similar products, so that children could express their preferences. The test–retest reliability was measured through the concordance and interclass co-relation between both answers. Results: Kappa concordance indices of 0.71 (0.46–0.97), 0.88 (0.62–1.15), and 0.90 (0.64–1.16) were obtained for the three foodstuff pairs. The interclass correlation coecient for the sum of manifested preferred scores was 0.72 (p = 0.0000). Conclusions: The paired selection method has been demonstrated to be ecient with good or very good reliability, making it useful for measuring the impact of food advertising on this age group.

relationship, such as the work of Halford et al. [17], which veri ed the ability of children exposed to advertising to recognize the advertised foods and showed that children who were obese recognized signi cantly more food advertisements than those who were not.
However, few studies have used experimental methodologies [12,18,19] to measure the causal relationship that might be established between exposure to advertisements and the consumption preferences of children; some have focused on measuring children's short-term preferences after controlled exposure [20][21][22][23][24]. In this line of work, the study by Borzekowski and Robinson [25] stands out.
In that study, in order to determine whether the display of food ads in uenced their preferences in the very short term, children were interviewed after being exposed to children's programs with commercial breaks in which index cards with pairs of products were shown, including the ones advertised, so that they could express their preferences, leading to the conclusion that children exposed to the ads were more likely to select the advertised foods.
This type of experiment, with a similar methodology, has also been used in other studies evaluating the effectiveness of advertising [22,[26][27][28], not necessarily of food products, both in child and adult populations, corroborating the usefulness of the method.
However, it is evident that preschool-age children (4 to 6 years old) are a group especially vulnerable to the presentation of consumption patterns in advertisements, and for this reason advertisers use speci c persuasive strategies aimed at these children [29][30][31]. Thus it is necessary to gain a better understanding of the in uence of these speci c strategies on consumer preferences at a stage of psycho-evolutionary development where many lifelong eating habits are acquired.
The objective of this paper is to evaluate the test-retest reliability of the paired selection method in the 4to-6-year-old age group, as a way to measure the outcomes of studies analyzing whether exposure to commercial advertising of food products exerts any effect on short-term consumption preferences at these ages.

Methods
A cross-sectional study was proposed, to evaluate the test-retest reliability of the paired selection method to measure the in uence of advertising on the participants' consumption preferences.
The study population was children 4 to 6 years old from the Escuela Infantil Santa Infancia in downtown Seville, which was randomly selected among the centers registered with the Ministry of Education of the Junta de Andalucía for the academic year 2011-12, which had authorized in that course 2 units of rst and second cycle of early childhood education (3-6 years) for 42 and 35 school posts, respectively (77 eligible subjects in total).
A convenience sampling was selected from 52 participants of these ages in the second year of elementary education, in attendance during June 2012.
As general demographic descriptive variables, sex and age were considered, and were used as independent variables in the analysis of factors associated with consumption preferences in the rst measurement. As dependent variables, the consumption preferences manifested by the participants after the use of the paired selection method were established.
The intervention consisted of exposing the group to an episode of the cartoon series "Caillou," trying not to make any reference to food consumption in its content. During the broadcast of the episode, a commercial break was included with 4 advertisements of various food products selected from ads studied in the work of Ponce-Blandón et al. [13], chosen by a group of experts using the following criteria: advertisements aimed at children, products regularly consumed by children with high fat content and/or re ned sugars, and/or that offer a gift and/or use the testimony of a famous person. Ads for these products were ultimately chosen: The child's response to each of the 4 advertised products was recorded on a data collection sheet. In case the child did not respond or made contradictory responses, it was marked as not valid and that participant was excluded from the study.
This procedure was applied 2 times (test-retest method) and by 2 pollsters instructed about the procedure, in order to evaluate interobserver reliability and repeatability [32]. The rst measurement was made at the end of exposure to the video by one pollster, and the second was made 15-30 minutes later by a second pollster. Both were unaware of the products advertised during the commercial breaks, so that they could not in uence the participants' preferences.
Subsequently, the data relative to the rest of the variables of the study were collected, facilitated by the participants´ tutor through the student center records.
All data in the data collection sheets were recorded with the Epi Info tool version 7.1.2.0 for further statistical analysis through the same application.
A descriptive analysis of the data was conducted, calculating the most representative frequencies of the qualitative variables, with con dence intervals of 95%. The quantitative variables were numerically summarized by calculating the main measures of central tendency and dispersion. In the same way, for the analysis of the relationships between sociodemographic variables and preferences in the rst measurement, the association between variables was estimated (odds ratio and relative risk, correspondingly) and the chi-square test was applied, establishing p-value < 0.05 as the level of signi cance.
With regard to the consistency analysis of the paired selection method (test-retest and interobserver reliability), the concordance between the 2 preference measurements for each product pair was studied.
To do this, we calculated the observed concordance ratio P o and the expected concordance ratio P e , as well as the kappa index and its 95% con dence interval. We established the following as reference values: ≤0.20 = very weak concordance, 0.21-0.40 = weak concordance, 0.41-0.60 = moderate concordance, 0.61-0.80 = good concordance, and ≥ 0.81 = very good concordance, accepting the outcome measurement tool for kappa index values above 0.60 (good or very good) [32].
The sum of the scores obtained by the preferences of each participant was calculated, with the preferences recoded as numerical variables with 2 possible values, 1 or 0, according to whether the preference was for the advertised product or not, generating a new variable for the sum of the scores of preferences expressed by participants, with values from 0 to 4. The interclass correlation coe cient was calculated to assess the numerical concordance between the rst and second measurement. Childhood Obesity of Andalusia. The right of the participants' parents not to expose their children to the experiment was preserved; they were provided with an informed consent document for their children's participation, and children whose parents did not sign the document were not included in the study.

Results
The results were obtained after the nal participation of 52 boys and girls out of the 77 eligible subjects, adjusted for losses, dropouts, and eliminations. The participants' ow through the study is represented in Fig. 1.
Regarding the ages, 42.3% (28.7-56.2) of the participants were 4 years old (n = 22), and 57.7% (41.3-69.5) were 5 or 6 years old (n = 30). The mean age was 4.6 years (SD = 0.5, (4.4-4.7)). In the rst measurement, 56.7% of preferences (42.9-69.1) were for advertised products, compared to 43% of preferences for the compared products (29.1-50.8). This difference did not turn out to be signi cant, as happened with the differences found in the preferences for each product pair shown separately, as seen in Table 2. Regarding the analysis of relationships between sociodemographic variables and preferences, no signi cant differences were found in preferences according to sex or the two age groups (4-year-olds and 5-and 6-year-olds), as can be seen in Table 3.  Table 4 shows the comparison of the two measurements of preferences by the paired selection method.   Figure 2 shows a bar graph comparing the frequencies obtained in the two measurements; in the rst measurement, 48.1% of the participants scored two points, and in the second measurement 34.6% obtained two points. The variance of the two measurements was 0.749 and 0.841, respectively, with an interclass correlation coe cient of 0.73 (p = 0.0000).

Discussion
The test-retest reliability of the paired selection method to measure the in uence of advertising on the food preferences of children 4 to 6 years old could be said to have been more than acceptable, taking into account that the kappa indices obtained for the four pairs of products between the two measurements were categorized as having good concordance for the rst two pairs and very good concordance for the last two pairs, and that the interclass correlation coe cient for the sum of scores for the two measurements showed good concordance [32], which were consistent when preferences were categorized as nominal variables and were converted to a numerical scale. Questionnaires were not used in this study, because the available tools were designed to be answered by parents of very young children (4 to 6 years old) and are more focused on measuring the frequency of eating habits. For example, the Food Frequency Questionnaire [36,37], the Children's Social Desirability for Food Scale [38], and other self-administered questionnaires such as the Youth Healthy Eating Index [39] and the one validated by Robinson et al. [26] are designed for older children rather than younger children, such as those in this study.
For populations with similar age characteristics and, more importantly, to measure their short-term consumption preferences, an extensively used technique is to measure the quantities of products consumed during the experiment or immediately after, offering participants various food products including those advertised in the commercial breaks of the experimental videos. This technique, used by different studies [22,24], seems to be effective at measuring the outcome, but it does have di culties, because it was necessary to lm and photograph the selections to ensure the correct measurement of the outcome [28] or their purpose was more to measure the quantities ingested of the products announced in the experiment [22,27].
The use of paired images of similar products, as in this study, was also employed by Reisch et al. [19], who showed 10 pairs of similar food products for which children expressed preferences by assigning an image of a smiley or sad face, indicative of desired or undesired products. It was also used with good internal and external validity by Kopelman et al. [34], who used paired cards with images of food products to check whether children exposed to advertisements were able to recognize logos and brands. Cornwell et al. [33] also successfully employed this technique, showing pairs of cards with images of fast foods to evaluate the recognition of brands, as did Borzekowski and Robinson [25], who handled pairs of cards showing sugary juices, donuts, sweets, chicken appetizers, sugary breakfast cereals, and peanut butters to assess the impact of advertising after 30 seconds in a group of preschoolers.
In almost all studies performing similar experiments, children have been exposed to a greater number of advertising messages and the outcome has been measured with a larger set of product pairs, but a methodological recommendation is mentioned to simplify the experimental procedure [19,25,33]. For this reason, in this paper it was decided to limit the pairs of products to four, in order to reduce the children's exposure to commercial messages, given their young age. Despite all efforts to avoid classi cation bias during the process, certain limitations must be assumed given the age characteristics of the population under study. In particular, a child's assignment of preference was sometimes di cult, since at this age it is very easy to in uence their opinion if the pollsters do not adopt an absolutely neutral stance, even in their nonverbal language. Thus, useful measures from other studies were adopted [25,33,34], such as pollsters having previous training to remain neutral, or the blindness technique, so that the pollsters' prior knowledge about the products could not in uence the children's responses. Even so, doubtful response from three children were considered and discarded from the de nitive sample, in order to guarantee maximum accuracy of the results.
Regarding the results analyzing the preferences and their differences according to the children's age and sex, it can be a rmed that the results were not conclusive. Two aspects cast shadows on this part of the experiment: the size and representativeness of the sample, and the absence of a real experimental methodological process, since there was no control group.
Indeed, the experimental method of exposing study subjects to children's programs with advertising breaks and randomly assigning participants to groups exposed to advertisements of food products versus control groups is common and proven to be able to evaluate the effect of advertising on children's consumption preferences. Including a control group that is shown the same animated video with breaks that include advertisements for nonfood products was done by some of the reviewed trials with similar hypotheses [17,20,40,41], with one control group shown advertisements of products such as toys, clothing, and other products, while another control group is shown the same video without ads. Regarding the ethical aspects, and in view of similar experiences [12,41], it is important to recognize that the intervention performed does not seem to pose any risk to the psychological integrity of the study subjects, since it does not go beyond what the studied population receives in terms of daily in uence. In fact, viewers of Andalusia´s main TV channels are exposed to almost seven ads for food products every hour and more than 30 general ads for each hour of programming, on average, and almost half of the ads for food and beverages are speci cally targeted to children [13].

Conclusions
In conclusion, measuring the in uence of advertising on short-term consumption preferences for processed food products by children aged 4 to 6 through the method of paired selection has proven to be effective, obtaining good or very good reliability for all measured preferences. Other experimental methodologies with the controlled and randomized exposure of children to advertising messages and the subsequent measurement of outcomes with paired selection could be recommended to contrast with the hypothesis of the association between exposure to advertising and short-term preferences.
This research will help develop measures to exercise control, of both the time children spend watching television and the persuasive and biased information often offered by advertisers, as well as to promote educational strategies for children to learn how to identify the more deceptive aspects of advertising in order to progressively choose healthier foods.

Declarations
Ethics approval and consent to participate The research was authorized by the University of Seville´s Ethics Committee of Experimentation. The right of the participants' parents not to expose their children to the experiment was preserved; they were provided with an informed consent document for their children's participation, and children whose parents did not sign the document were not included in the study.

Consent for publication
Not applicable.
Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
42. Gregori D, Ballali S, Vecchio MG, Sciré AS, Foltran F, Berchialla P. Randomized controlled trials evaluating effect of television advertising on food intake in children: Why such a sensitive topic is lacking top-level evidence? Ecol Food Nutr. 2014;53:562-77. Figure 1 Flowchart of participants ow through the study.