Global Warming’s “Six Americas Short Survey”: Audience Segmentation of Climate Change Views Using a Four Question Instrument

TLDR

Audience segmentation, long used in marketing, health, and communication, is now a key tool in environmental research, exemplified by the Six Americas model that classifies Americans on climate change beliefs but requires a 36‑question screener that limits its practical use. This study develops the Six Americas Short Survey (SASSY), a four‑question screener derived from 14 national samples and machine learning that accurately assigns respondents to the six climate‑change segments, and offers a web application to support outreach. Using machine‑learning algorithms on 14 national survey datasets, the authors selected four items measuring risk perception, worry, expected harm to future generations, and personal importance of global warming to construct the SASSY screener. SASSY attains high true‑positive accuracy across all six segments on a 20‑hold‑out set, replicates these results in four out‑of‑sample validations, and demonstrates test‑retest reliability in an independent two‑wave sample.

Abstract

Audience segmentation has long been used in marketing, public health, and communication, and is now becoming an important tool in the environmental domain as well. Global Warming's Six Americas is a well-established segmentation of Americans based on their climate change beliefs, attitudes, and behaviors. The original Six Americas model requires a 36 question-screener and although there is increasing interest in using these segments to guide education and outreach efforts, the number of survey items required is a deterrent. Using 14 national samples and machine learning algorithms, we identify a subset of four questions from the original 36, the Six Americas Short SurveY (SASSY), that accurately segment survey respondents into the Six Americas categories. The four items cover respondents' global warming risk perceptions, worry, expected harm to future generations, and personal importance of the issue. The true positive accuracy rate for the model ranges between and across the six segments on a 20 hold-out set. Similar results were achieved with four out-of-sample validation data sets. In addition, the screener showed test-retest reliability on an independent, two-wave sample. To facilitate further research and outreach, we provide a web-based application of the new short-screener.