A Weighted Multidimensional Index of Child Well-Being Which Incorporates Children’s Individual Perceptions

A thorough analysis of the several methodologies is provided in Fernandes et al. (2012).

It should be noted that none of the abovementioned studies consider interactions between the dimensions of well-being that, according to Bronfenbrenner and Morris (1998), are likely to exist. Bronfenbrenner and Morris’s (1998) ecological model of human development considers that the main effects on children’s outcomes are likely to be the result of interactions between factors. Although comprising a limitation to the existing indexes in this area, developing an interaction model between dimensions of well-being is beyond the (necessarily restricted) scope of the present research work. Nevertheless, some comments on this matter are made in the concluding section.

Full references are given in Bronfenbrenner and Morris (1998).

Questionnaires are available upon request to the corresponding author.

Income categories in the questionnaire were defined according to the Portuguese National Statistics Institute’s (INE 2008) categorization of the annual total net income of households, divided by the twelve months of the year in order to obtain monthly income.

Moore et al. (2007, 2008) also refer to these components as ‘risk factors’ and analyze them together with contextual variables instead of within the group of individual well-being variables.

Two remarks should be made about the overall happiness question, one about the wording and another about the scaling. First, the word “happiness” was used instead of “satisfaction” because “happiness” seems to be a much easier concept for children to understand as the term “satisfaction” is not very commonly used amongst children and young people (Rees et al. 2010). Second, a ten-point scale was used to assess overall happiness because, following Cummins’ (2003) work on life satisfaction scales, these types of scales have demonstrated to be superior in representing people’s levels of life satisfaction when compared to smaller scales of five or seven points.

For other works on alternative weighting schemes see, for example, Decancq and Lugo (2010), Guio et al. (2009), and De Kruijk and Rutten (2007).

For some more definitions see, for example, Diener (1994).

According to Oppenheim (1992), when measuring attitudes it is preferable not to have many neutral items or many extreme items at either end of the scale, which justifies the choice of a five point scale.

According to the Portuguese Ministry of Education (www.​gepe.​min-edu.​pt), in 2008/2009 and for the North region of the country, around 183,500 children were registered in the third, fourth, fifth and sixth grades of the education system.

Data was gathered through two questionnaires applied to children themselves and their parents/carers. The questionnaires were sent to the schools that agreed to take part in the study with precise instructions on how to be applied. Children answered the questionnaire in the classroom and the parents’ questionnaire was sent home to be completed and then returned. Only one of the parents/carers had to answer the questionnaire that was sent home. The process of data gathering took place between April and June of 2011.

The exclusion of cases was made based on inconsistencies between the children’s and parents’ answers concerning the existence of siblings/other children in the household. For all other cases where the inconsistency related to non-absolute answers, these were corrected: for the cases where data was missing on the children’s behalf, the parents’ answer was considered; and for the cases where data was missing on the parents’ behalf, the children’s answer was considered.

The original parents’ questionnaire also included a question about the household’s ethnic origins but due to the large amount of missing data, this information was considered unsuitable for analysis.

Regarding the objective individual indicators (reported by parents), two rules of thumb were considered: “having” (for positive indicators, “not having” for negative indicators) is better than “not having” (“having” for negative indicators) and “having more” is better than “having less”. So, considering a positive indicator of any one of the dimensions defined, we assume for every individual child that if he/she has a certain item (scoring 1) then that same child is better off than a counterpart that does not have that item (who scores 0). Following the same type of reasoning, if an individual has more of a certain item (scores more than 1), that individual will be better off than another one who has the item but in a lesser quantity. Additionally, we also consider the importance degree the child states each item to have for him/her. Thus, if a child has a certain item, his/her well-being will increase according to the importance he/she attaches to that item. For example, if the child has an item (scores 1 in objective terms) and on a scale from 1 to 5 values that item in 4, then his/her well-being score for that item will be 4; if a child does not attribute any importance to that item (reports 1 in terms of importance degree), then his/her well-being score will be 1. On the other hand, if the child does not have that item, she/he will be attributed a negative score according to the importance degree she associates to the item, plus 1 (so negative scores can go from -4 to -1). Also according to this, if the child does not have the item and does not value that item then his/her well-being score will be 0.

We imposed that within dimensions all the weights have to sum one. To compute the overall well-being index, the individual domain indexes were summed using the same procedure, having each of the well-being dimensions been attributed a different weight, which also resulted from children’s ranking of dimensions.

Given that the index comprises so many items, it was very likely that any given child might be missing an item, which ended up having impact on the overall sample size. So for analysis purposes and in order to maximize the sample size we imputed missing data. Data was imputed considering the most common responses (modal response) according to selected features of the household and/or the child (income category, sex, age, school year, having siblings, etc.) and this method was adopted for imputation of missing data concerning parents’ answers and for children’s answers regarding the degrees of importance. In what respects the weights of each indicator and dimension attributed by children, missing data was imputed using the mean weights, also determined according to a set of child features (sex, age and school year). We chose not to impute missing data for income nor for other variables characterizing the household and the child (nationality, mother and father’s education level and employment situation, type of school frequented by the child, child’s age, gender and school grade, reported happiness degree by parents and by the child).

Although for our sample gender does not reveal to be significant, it is interesting to note the change of direction of the impact on overall well-being when weights are introduced in the calculation of the index (compare Scenarios 1 and 2 with Scenario 3 and our proposed index—Table 5). When no weights are considered, female children seem to do better than their male counterparts, but when introducing weights, females do worse than males.