Data were collected in August 2014 in Australia, Germany, Sweden, UK, and US. A web-based questionnaire was distributed by an Internet survey company to respondents representative of the adult population in each of these five countries in terms of age, sex, and educational level. Per country at least 500 respondents were recruited for the study. This number was motivated by efficiency measures of the design used in this choice experiment . The number of respondents included in the data analysis was: 551 (Australia), 562 (Germany), 548 (Sweden), 552 (UK), and 550 (US). Descriptive statistics of the study samples can be found in Table 1.
Discrete Choice Experiment
Discrete choice experiments were conducted to develop tariff sets for the CarerQol-7D for Australia, Germany, Sweden, UK, and US. The methodology applied here is based on the estimation of the CarerQol tariff for The Netherlands . Respondents were instructed to imagine that while completing the survey they provided care or support to a loved one as a result of an illness, disability, or infirmity of old age. They were also asked to keep the same care recipient in mind during the whole experiment. After completing the choice tasks, respondents were asked for the level of difficulty of the choice tasks and their familiarity with informal care giving, or with caregivers in their own circles of family and friends.
Choice sets were constructed with an efficient experimental design with priors from the CarerQol-7D tariff set for the Netherlands  to calculate standards errors of the parameters as statistically efficient as possible to increase reliability of the results with smaller sample sizes [27–29]. The efficient experimental design contained 40 choice sets,Footnote 1 which were blocked over four groups of respondents, i.e., in the survey, ten choice sets were presented to each respondent. The utility functions consisted of two dummy variables per attribute (reference level: no for the two positive attributes of the CarerQol-7D, a lot for the five negative attributes), interaction terms for all attribute combinations, and a constant term for the first alternative in the choice set. The dummy variables were treated as Bayesian priors with a normal distribution using mean and standard error of the multinomial logit (MNL) model of the tariff for the CarerQol-7D in the Netherlands, allowing parameter values to be both negative and positive (see “Appendix”, Table 5). The efficient experimental design was optimized for D-efficiency in the basic multinomial logit model  calculating mean values using 1000 Halton draws . The efficient experimental design was constructed in Ngene (ChoiceMetrics, Australia).
In this discrete choice experiment, respondents were asked to choose between two unlabeled hypothetical informal care situations (see Fig. 2 for an example). These hypothetical informal care situations were described by a combination of the seven attributes and three levels of the CarerQol-7D: (1) fulfilment with carrying out your care tasks, (2) relational problems with the care receiver, (3) problems with your own mental health, (4) problems combining your care tasks with your daily activities, (5) financial problems because of your care tasks, (6) support with carrying out your care tasks, and (7) problems with your own physical health. The levels of these attributes were ‘no’, ‘some’, and ‘a lot’. Color coding was used to aid visual representation of the information: positive attribute levels were displayed in green text, the negative levels in red text, and the intermediate levels in orange text . The choice sets were presented in random order to the respondents.
In addition, at the start of the questionnaire, information was gathered on the respondents’ age, sex, highest attained educational level, marital status, and current employment status.
The questionnaire was translated in Swedish by native speakers involved in research among informal caregivers and in German by Mapi values. The original English version of the CarerQol instrument  was translated into Swedish and German using forward–backward translation. The English translation was performed by the authors and checked for accuracy by native speakers and informal care researchers from Australia, UK, and US. This resulted in separate questionnaires for each of the three countries, which were largely identical but contained some country-specific adaptations to spelling and wording (e.g., ‘neighbours’ or ‘neighbors’ in the examples provided with the seven dimensions of the CarerQol).
The data were analyzed with a panel mixed multinomial logit (MMNL) model, allowing for the presence of unobservable preference heterogeneity in the population (random parameters) and the correlation of responses across observations (panel structure) [26, 32]. Likelihood ratio tests were used to construct the model specification, in particular considering whether the model should include: (1) an alternative specific constant, (2) random or fixed parameters, (3) interaction effects of the attributes, and (4) collapsed attribute levels. These tests were performed for each country separately.
The tariff for the CarerQol-7D was based on individual-specific parameters. These parameters were calculated by randomly assigning the unconditional distribution of the panel MMNL model (population level estimates of the panel MMNL model) over a hypothetical sample of 10,000 individuals with bootstrap sampling . Next, the individual-specific parameter estimates from the bootstrap sampling were averaged. The mean parameter values were rescaled to represent the CarerQol-7D tariff ranging from 0 (worst informal care situation) to 100 (best informal care situation). The standard errors of the tariff were calculated by dividing the standard errors of the MMNL parameters by the same total score. Analyses were performed in Nlogit 5.0 (Econometric Software Inc., Plainview, New York, US). Finally, for the purpose of illustration, CarerQol-7D utility scores were calculated for six caregiving situations described by the CarerQol-7D using the tariffs of Australia, Germany, Sweden, UK, and US. These caregiving states were selected given their prevalence in a large dataset of informal caregivers in the Netherlands . The six CarerQol-7D states ranged from less to more desirable caregiving situations (see Table 4). The CarerQol-7D utility scores of the five countries of this study and of the Netherlands, computed using the tariffs presented in Hoefman et al. , are presented in a radar plot created in Microsoft Excel (US).