Comparing survey and transaction data

A survey records answers from eligible respondents. Transaction data records purchases or payments inside a defined system.

To compare their findings, first decide whether they describe the same people, products and period. Separate aggregate headlines cannot show that the same shoppers said one thing and did another.

Identify the observation in each data set

A survey can answer a question about a reported reason, preference, difficulty or intention. Check who was invited, who responded, the exact question, its answer options and the field dates.

A transaction extract can describe recorded amounts, items and timing. Check the included merchants or stores, channels, payment methods, returns and event dates. It usually cannot tell you why a purchase occurred.

Neither data type is automatically more representative. A large transaction file may omit customers or payment routes; a survey can have coverage, non-response or response error. The relevant risks depend on how the particular data were collected.

Choose the comparison you can make

If the sources describe different people, report two bounded findings side by side. Align the geography, product definition and dates where possible, but do not call any difference a gap between those people’s words and actions.

If a retailer links consenting respondents to their later purchases, define the eligible group and follow-up window. Check how many respondents can be linked, which purchases remain unseen and whether a response preceded the recorded action. Compare the same product or service choice on both sides. Even a linked record does not, by itself, establish that the stated reason caused the purchase.

For an aggregate comparison, also check the measure. A survey share of respondents naming price cannot be subtracted from a change in transaction dollars. Prices, quantities and product mix can all affect the latter.

Recognise a public-data limit

The ABS Monthly Household Spending Indicator uses aggregated, de-identified bank card transaction data provided to the ABS by participating banks. It also uses supermarket transaction files sent weekly and new vehicle sales data. These inputs are aggregate and transaction-file based, not linked respondent records. For this comparison, the published spending estimates are not designed to be linked to an individual survey respondent’s later purchases.

Report each conclusion with its own verb: respondents reported a view; transactions recorded in-scope activity. A claim about the same people’s behaviour needs an appropriately linked and sufficiently complete observation.

Key Facts on ABS Monthly Household Spending Indicator