RDP 2018-11: Consumer Credit Card Choice: Costs, Benefits and Behavioural Biases 3. Data

The data used in this paper come from two sources: the RBA's 2016 Consumer Payments Survey, and information on advertised credit card features collected from card issuers' websites.

The 2016 wave of the Consumer Payments Survey contains data on 1,510 respondents representative of the Australian adult population, 72 per cent of whom held a credit card.^[11] The survey includes information on respondents' demographic characteristics, card ownership and use of various payment methods as logged in a week-long diary. Full details of the survey methodology are available in Doyle et al (2017). Throughout most of this paper, I limit analysis to respondents who owned a credit card, reducing the sample to around 1,080 respondents.^[12] I apply survey weights to all regressions, tables and graph data so that the final (reweighted) dataset aligns with demographic benchmarks for the Australian population.

This paper relies on a range of questions that were added to the Consumer Payments Survey in 2016. Questions were asked that were relevant to the behavioural hypotheses outlined above: on the features respondents considered most important when they chose their card, their perception of the monetary value of their card, and whether they signed up to their card with a temporary special offer. The survey also asked whether respondents had switched, or considered switching, cards in the past year.

To facilitate the calculation of the net monetary benefit, respondents were asked about how they use their main credit card. The survey asked about how respondents redeem their rewards points, whether they had paid a discounted annual fee, their usual card repayment behaviour, and which alternative source of credit they would use, if any, if they did not have a credit card. A copy of the relevant sections of the questionnaire is in Appendix F.

In addition to the survey data, I use data on the features of the particular credit cards held by respondents. As part of the Consumer Payments Survey, each respondent identified their main credit card (if they held multiple cards, this was the one they used most often). The detailed features of these cards, including fees, interest rates and rewards programs were obtained from card issuers' websites in late 2016. These data are collected by the RBA for internal analysis, though a range of comparison websites provide similar information.

Much of the information used in this analysis is self-reported. In many respects, self-reported information is valuable. We were able to ask respondents about their preferences and about which factors they considered to be important when they chose their credit card. But self-reported information on financial behaviours – such as frequency of card repayments, total monthly spending and monthly interest charges – may be subject to two potential response biases. First, there is a possibility of misreporting due to social desirability bias, with some evidence suggesting consumers are more likely to respond to surveys in a way that reflects favourably on their financial management skills (Kelly et al 2017). For instance, respondents may state that they ‘always pay off their balance in full’, even if this is not the case. This would lead to an underestimate of the value of interest charges. To minimise this potential bias, respondents were asked to refer to their most recent credit card statement in answering the questionnaire. Second, reliance on self-reported information reduces the sample size, as some respondents did not answer all questions, and others did not enter meaningful information in some fields.^[13] As a result, I am unable to calculate the net benefit for around 20 per cent of credit card holders (after applying population weights), though dropping these respondents does not appear to meaningfully affect the results.^[14] As an additional check on the representativeness of my sample, I compare components of the net monetary benefit with estimates from other sources in Section 4.3.

Footnotes

Due to different response rates across demographic categories, survey weights are used so that the final (weighted) dataset aligns with population benchmarks, one of which is credit card ownership. After weighting, 54 per cent of survey respondents held a credit card. [11]

I also exclude 13 respondents who held a credit card with an annual fee above $1,000, as I was unable to quantify the value of certain non-reward points benefits that these cards carry. [12]

For instance, respondents were asked to select their credit cards from a list and, if they could not find their card on the list, to enter the name of their card as free text. I was able to reclassify most free text responses, but 4.5 per cent of cardholders entered free text from which I was unable to identify the particular card. [13]

Overall, the characteristics of respondents with missing data are similar to the sample average, though they tended to be younger and to have lower incomes. They were also a little more likely to report that they regularly pay interest charges, and generally they held less positive attitudes towards credit cards than other respondents. If anything, dropping these respondents is likely to upwardly bias my estimates of the net monetary benefit. Appendix B presents sensitivity tests that calculate the net monetary benefit using alternative variables that are available for more respondents. These changes increase the sample size, but do not meaningfully affect the distribution of net monetary benefits. [14]