IHP says General Election voting intent estimates uncertain due to substantial response bias

Thursday, 14 November 2024 00:57 - - {{hitsCtrl.values.hits}}

SLOTS polling of voter intent conducted by the Institute for Health Policy (IHP) since the Presidential Election through 8 November has suffered from an increased level of response bias.

“This has been due to respondents’ over-reporting support for the NPP. This has resulted in large uncertainty and a significant bias in favour of the NPP in the resulting vote intent estimates,” said IHP in a statement.

“Nonetheless, we are releasing estimates based on these polling data in the interests of transparency. We advise that these estimates are likely to be associated with a large margin of error, and caution should be taken when interpreting them. As with any opinion poll, they should not be regarded as a prediction of the final election result,” it added.

Starting in about June 2024, the IHP SLOTS poll experienced increasing bias in survey responses. The share of respondents claiming to have voted Anura Dissanayake in 2019 increased from 2-3% in the previous two years of polling to 10-15%, compared with the actual election result of 3%. Since the Presidential Election on 21 September, this trend has been reinforced, with over 60% of respondents claiming they voted for Anura Dissanayake in this year’s election, and additionally claiming intent to vote for the NPP at the forthcoming General Election.

These numbers are inconsistent with the actual election results and point to substantial bias in our polling due to significant social desirability bias favouring the NPP. This bias is likely to result in over-estimate of support for the NPP. Although IHP takes steps to counter bias in its analysis, it is not possible to fully do so as the level of bias in this case is unknown.

“Our best analysis of the SLOTS survey data estimates support at end – October/early November for the NPP at 53% of all adults, followed by the SJB with 26%, NDF with 9%, SLPP with 7%, and the ITAK with 2%. These figures are subject to considerable uncertainty and a subjectively assessed margin of error of 5–7%. This error may be even larger for the reasons mentioned above. Internal modelling of these estimates indicates this would translate into 120-130 seats for the NPP, followed by the SJB with 55-65 seats. Because of the bias in our data, we cannot exclude the possibility that NPP support may be higher than implied above. In that scenario, the NPP may win a two-thirds majority in Parliament, but we have less confidence in this outcome,” IHP said.

IHP’s SLOTS MRP methodology first estimates the relationship between a wide variety of characteristics about respondents and their opinions – in this case, “If there was a General Election today, who would you vote for?” – in a multilevel statistical model that also smooths month to month changes. It then uses a large data file calibrated to the national population to predict voting intent in each month since October 2021, based on what the multilevel model indicates about their probability of voting for various parties (Post-stratification) at each point in time.

IHP’s data collection covers all districts and communities, and it uses raking to ensure that the estimation samples fairly represent the adult population with respect to age, gender, ethnicity, religion, province and district, sector of residence, relative socioeconomic status, and education.