The Effect of Language on Political Appeal: Results from a Survey Experiment in Thailand

Distinctions should be drawn here between the concepts of language, dialect, register, and accent. First, for the division between language and dialect, I use the coding produced by Lewis et al. (2015), which is based on mutual intelligibility. If speakers are able to understand each other at a functional level without need to learn the other tongue, then the difference is considered one of dialect. Otherwise it is a language distinction. Second, linguistic registers are stylistic variations within a single language based on social contexts. Some sociolinguists distinguish between the concepts of style and register, with style being associated with level of formality and register being specifically focused on vocabulary (Trudgill 2000). This distinction, though, is relatively recent, and historically style and register have been mixed (see Joos 1967). As there is heavy conceptual overlap between formality and specific vocabulary, especially in Thai, I primarily use the term register. Finally, accents are differences in pronunciation of a single language, and they are often associated with social class but they are distinct from register (Trudgill 2000). A speaker can use a single accent to speak in a high register or low register, or even different languages, although certain accents are generally associated with lower registers. As such, I treat accent as distinct from register, and I control for accent in the treatment below.

Ethnic appeals do not necessarily guarantee political support (Carlson 2015; Dunning and Nilekani 2013; Horowitz and Klaus 2018).

Standard Thai is the official, codified version of the Thai language and is often described as Central Thai. For a discussion of language categorization in Thailand see Diller (2002, pp. 76–80).

Respondents self-identified the language spoken at home as follows: (1) Isan, 653 individuals (87.07%); (2) Central Thai, 60 individuals (8.00%); (3) Cambodian, 27 individuals (3.60%); (4) Lao, 6 individuals (0.80%); (4) Other, 4 individuals (0.53%). 664 respondents (88.5%) speak Isan at either home or work. The analysis below was repeated with data from only those who speak Isan, with similar results. That data is reported in Online Appendix.

The sampling procedure proceeded as follows. (1) For each of the three provinces in the survey, I randomly selected a list of ten districts (amphoe), with the districts being weighted according to district populations in the provinces. Each of those ten districts was assigned twenty-five surveys. (2) Within each of the selected districts, I randomly selected two sub-districts (tambol), with sub-districts being weighted according to their populations. Each of these sub-districts was then assigned either twelve or thirteen surveys. (3) The survey team travelled to each of these assigned sub-districts, arriving in the same village as the sub-district office. There the team would begin at a residential street, and using a skip number of three, choose homes to approach for the interview. If no suitable respondent was available, the team would move to the next house. In each sub-district, the survey continued until the target number of respondents was met.

252 Central Bureaucratic Thai treatments, 250 Central Thai treatments, and 248 Isan treatments.

A full version of the survey can be found in the online appendix. There is some concern that the treatment’s effect may have dissipated over the time that it took to gather responses. Despite this, we can be reasonably confident that the effect of the 2-minute treatment lasted during the brief response window (approximately ten minutes). Also, difference of mean tests for the treatments across the last two statements were statistically significant. Thus, even if the impact of the treatment diminished slightly during the survey, responses continued to exhibit an effect.

Demographic differences across provinces, though, were sufficient to warrant concerns that provincial effects may influence findings. As such, I repeated the analysis using ordered logistic regression models with provincial fixed effects. In addition, I conducted a separate analysis using a binary logistic regression model to obtain marginal effects. Results were similar to those presented here and are provided in Online Appendix.

Results of the factor analysis can be found in Online Appendix.

There were 463 qualitative responses, or 61.7% of the surveys.

Only seven respondents specifically highlighted language in their comments; another four mentioned manner of speech.