Citizens form expectations about election outcomes to make voting decisions. Surveying individuals for these vote expectations is also used to forecast elections. While individual expectations might not be as accurate, aggregating them can lead to accurate citizen forecasts. Unlike traditional vote intention polls, representative or large samples are not required for citizen forecasting. It is thus a promising approach for forecasting local or constituency elections where forecasts to date remain scarce.
Yet, for now success rates of citizen forecasting vary. This could be due to varying individual forecasting competence and having to forecast in different contexts. In plurality systems such as the United States or the United Kingdom, citizen forecasting has been very successful. In contrast, in more complex systems, like in Germany's mixed proportional representation system, findings have been more ambiguous. Moreover, previous research has shown that citizens vary in their forecasting competence. For now, there is, however, no clear-cut answer to which citizens are more accurate forecasters and under which circumstances this is the case. Therefore, I synthesize previous research findings through a meta-analysis to answer the following question: to what extent do individual and contextual characteristics affect the accuracy of citizen forecasts? Individual characteristics include political preferences, political knowledge, poll consumption, social networks, and demographics. Additionally, contextual characteristics comprise differences across studies regarding the elections, i.e., tightness of the race, level of the election, or electoral system, and regarding survey details, i.e., the phrasing of the forecasting task or the survey timing in relation to election day.