Polls in Non-Democratic Countries

When Honesty Is Dangerous

Polling in democracies assumes that respondents will answer honestly, protected by anonymity. In authoritarian regimes, this assumption fails. Citizens who fear government surveillance may give the 'safe' answer rather than their true opinion. This is called 'preference falsification,' a term coined by economist Timur Kuran.

In the Soviet Union, surveys consistently showed overwhelming support for the Communist Party, right up until the system collapsed in 1991. In Iran, pre-election polls showed strong support for the regime's preferred candidates. In China, surveys consistently show high satisfaction with the central government. Whether these results reflect genuine opinion or strategic self-censorship is one of the hardest questions in political science.

Measuring the Unmeasurable

Researchers have developed creative techniques to pierce the veil of preference falsification. List experiments embed sensitive questions within lists of innocuous items, allowing statistical estimation of the sensitive opinion without any individual revealing their view. Implicit association tests measure unconscious attitudes. Analysis of social media behavior and economic choices (voting with your feet through emigration) provides indirect evidence of true preferences.

International polling organizations like Pew and Gallup conduct surveys in authoritarian countries but interpret results cautiously, noting the context in which responses are given. A 90% approval rating for the government in a country where dissent is punished tells us something, but what it tells us is ambiguous.

A Worked Example: The List Experiment

Suppose you want to know what share of citizens in an authoritarian state privately oppose the ruling party — a question almost no one will answer directly. A list experiment sidesteps the fear. You randomly split 2,000 respondents into two equal groups. The control group (1,000 people) sees three uncontroversial statements and reports only HOW MANY they agree with, never which ones:

• The economy affects my family.
• Traffic in my city is bad.
• I follow the national football team.

The treatment group (1,000 people) sees the same three items plus a fourth, sensitive one: 'I oppose the ruling party.' They also report only a count. Because no individual ever points to the sensitive item, no one is exposed.

The difference in means — 2.32 − 1.80 = 0.52 — estimates that about 52% of respondents privately oppose the ruling party, even though a direct question might show 90% support. The technique trades precision for safety: the estimate carries a wider margin of error than a direct question would, and it cannot tell you WHICH individuals hold the view. But it converts an unanswerable question into a measurable one. Researchers have used variants of this design to estimate hidden support for insurgents and hidden opposition in single-party states — precisely because the Soviet lesson (near-unanimous approval right up to the 1991 collapse) taught that direct questions in fearful societies measure the regime's reach, not the public's mind.