RP/SP consistency checks in transport DCE studies

In combined RP/SP transport studies, inconsistencies between revealed and stated preference data reveal respondents who are not engaging genuinely with the stated preference tasks.

This article explains RP/SP consistency checking in combined transport surveys - what consistency means, how to detect violations, and how to handle them in analysis.

Ben White, 07.07.2026

What is RP/SP consistency?

In a combined RP/SP transport survey, each respondent reports their actual travel behaviour (revealed preference) and then makes choices in hypothetical scenarios (stated preference). If a respondent reports that they drive because it is cheaper than the train, but in the SP tasks they consistently choose train alternatives even when they are more expensive than the car alternatives, this is an RP/SP inconsistency.

RP/SP inconsistencies can arise for legitimate reasons - the SP scenarios might offer a significantly improved train service that the respondent genuinely prefers to their current driving experience. But systematic inconsistencies - particularly those that defy basic economic logic - are a data quality concern.

Why RP/SP consistency matters for VoTT estimation

Value of travel time estimates from joint RP/SP models are sensitive to the consistency between the two data sources. If SP choices are inconsistent with RP behaviour, the joint model cannot reconcile the two datasets effectively, and the resulting parameter estimates may be unreliable.

The scale factor in joint RP/SP models - which accounts for the difference in variance between RP and SP error terms - is only identifiable if the two datasets are genuinely from the same underlying preference structure. Systematic RP/SP inconsistencies violate this assumption.

SurveyEngine's combined survey export provides the data needed for consistency checking. The RP mode, RP time, and RP cost can be compared against the SP choices to identify implausible response patterns.

TLDR Quick links

Transport resources

Fraud detection toolkit

Open a free account

Checking RP/SP consistency in SurveyEngine data

Step 1: Export the combined RP/SP dataset from SurveyEngine. The export includes each respondent's RP mode, time, cost, and SP choices with the attribute levels they saw (including the pivot-adjusted values).

Step 2: Check whether RP mode is dominated in the SP tasks. For each respondent, identify SP choice sets where an alternative equivalent to their RP mode (same time and cost as their actual trip) was available. A respondent who consistently rejects their RP mode equivalent in SP, even when the SP alternative offers no advantage, may not be engaging genuinely.

Step 3: Check for implausible preference reversals. Compare the implicit VoTT implied by each respondent's SP choices against expected ranges. Respondents with implausibly high or negative VoTT may be applying non-compensatory heuristics rather than genuine trade-off reasoning.

Step 4: Flag and review implausible respondents. Create a flag for respondents who fail the consistency check. Review their SP responses and RP data manually to determine whether the inconsistency has a plausible explanation or represents low-quality data.

Step 5: Run sensitivity analyses. Estimate joint RP/SP models with and without consistency-flagged respondents. If VoTT estimates change substantially, investigate the source of inconsistency further before finalising your estimates.

Worked example - commuter mode choice consistency

A commuter mode choice study finds 34 respondents (8.5%) whose SP choices are inconsistent with their RP mode. Manual review identifies three groups: 12 respondents who are car users but consistently choose public transport in SP regardless of relative cost (plausible - latent demand for PT), 14 respondents whose SP choices imply a VoTT below zero (implausible - suspected straight-lining), and 8 respondents whose SP choices appear random (implausible - suspected inattention).

The 22 implausible respondents are excluded from the main analysis. The 12 consistent latent PT users are retained and their higher PT preference is captured in the model through an alternative-specific constant for public transport.