Polling errors on California tax measures raise questions about data accuracy

A review of California ballot measure history reveals that polls often misjudged voter sentiment on tax reform issues by wide margins.

The Howard Jarvis Taxpayers Association, which sponsored several of these measures, noted that polling data consistently undercounted support for taxpayer interests. Jon Coupal, the association's president, stated that while polls are indispensable for measuring sentiment, they face challenges in a state like California when measuring support for conservative fiscal causes.

The most dramatic discrepancy occurred during the 1996 vote on Proposition 218. Leading up to the November 5 election, polls showed the measure failing badly. The final Field Poll released before the vote indicated only 36% of likely voters supported the proposal. In the actual result, the measure passed with 56.55% support, defying negative polling predictions by a gap of over 20 points.

Earlier, Proposition 13 in 1978 showed a similar pattern of shifting support. Shortly after Howard Jarvis filed 1.2 million signatures, early polls showed voters divided, with support at 25% and opposition at 27%. By spring, as property tax bills arrived, support rose. However, the last Field Poll before the June 6 primary projected passage at roughly 57% to 60%. The final margin was 64.8% to 35.2%, meaning the poll understated support by several points.

More recent data from Proposition 5 also showed inaccuracies. This measure sought to lower the vote threshold for local bonds from two-thirds to 55%. A Public Policy Institute of California poll released just before the election reported a nearly even split, with 49% supporting and 50% opposing the amendment. The measure failed by a margin of over 55% to 45%, placing the poll's error at approximately 5%.

Nationally, concerns about polling reliability have increased following the collapse of Wedgewood Polls. The firm admitted its data was entirely fabricated despite being included in major aggregators such as The New York Times, FiftyPlusOne, and VoteHub. The National News Desk reported that the firm lacked history, formal credentials, or real data behind the vast majority of its releases.

Another high-profile error occurred in Iowa in 2024. A Des Moines Register poll conducted by J. Ann Selzer showed Kamala Harris leading Donald Trump by three points. Trump ultimately won the state by 13.2 points. Following the result, Selzer retired from polling, and the Des Moines Register's reputation took a hit. Coupal argued that these discrepancies suggest many voters conceal their fiscal conservatism or do not respond to polls, making election day the only definitive measure of voter intent.