Uncovering the Nexus Between Attitudes, Preferences, and Behavior in Sociological Applications of Stated Choice Experiments [original]

Article

Uncovering the Nexus

Between Attitudes,

Preferences, and Behavior

in Sociological Applications

of Stated Choice Experiments

Ulf Liebe

, Petr Mariel

, Heiko Beyer

and Ju

¨rgen Meyerhoff

Abstract

Multifactorial survey experiments such as stated choice experiments are

used more and more frequently in social science research. In this article,

based on an experimental study on ethical and political consumption, we

explore the potential of hybrid choice models to explicitly model latent

psychological factors such as attitudes, overcoming a possible endogeneity

bias and misrepresentation of causality. To this end, we employ a hybrid

latent class choice model (HLCCM) in which the latent class structure

allocates individuals to classes according to underlying latent attitudes that

also influence the answers to attitudinal questions. This allows, in line with

sociological action theories, a theory-guided testing of preference segmen-

tation and modification caused by attitudes. We compare the complex

Department of Sociology, University of Warwick, Coventry, United Kingdom

Department of Applied Economics III (Econometrics and Statistics), University of the Basque

Country, Leioa, Spain

Institute for the Social Sciences, Heinrich Heine University Du¨sseldorf, Du¨sseldorf, Germany

Institute for Landscape Architecture and Environmental Planning, Technische Universita

¨t

Berlin, Berlin, Germany

Corresponding Author:

Ulf Liebe, Department of Sociology, University of Warwick, Coventry CV4 7AL, United

Kingdom.

Email: [email protected]

Sociological Methods & Research

2021, Vol. 50(1) 310-347

ªThe Author(s) 2018

Article reuse guidelines:

sagepub.com/journals-permissions

DOI: 10.1177/0049124118782536

journals.sagepub.com/home/smr

HLCCM with less complex models that do not take the latent variable nature

of attitudes into account and discuss in which cases less complex models

might be more appropriate. However, the HLCCM always has the advantage

of providing structure for theory testing and is therefore a useful tool to

uncover preference heterogeneity, preference modification, and decision-

making processes in sociological and other social science research.

Keywords

attitudes, hybrid choice model, latent class analysis, stated choice experi-

ment, stated preferences

While factorial surveys have been widely used in sociological research since

decades (see Wallander 2009 for an overview), other multifactorial survey

designs such as stated choice and conjoint experiments are still novel for

most sociologists (but see Auspurg and Hinz 2014, 2015; Beyer and Liebe

2015; Liebe et al. 2016). In stated choice experiments (SCEs), respondents

are asked to choose from an array of behavioral alternatives, which vary in a

number of attributes, the alternative they favor most. This design allows

researchers to estimate the effect of each attribute on respondents’ stated

choices. SCEs originated in marketing and transportation economics (Lou-

viere, Hensher, and Swait 2000) and became popular in many subfields of

economics including transportation, health, and environmental economics

because they provide a means of measuring preferences for product attributes

even if the good in question is hypothetical. It is important to stress that SCEs

are not identical with or a special case of conjoint experiments. The main

difference is the theoretical foundation of SCE, which is based on random

utility theory (see Auspurg and Hinz 2015; Louviere, Flynn, and Carson 2010

for a detailed discussion). This theoretical foundation—a rational-choice

framework—is in-principle in line with many sociological action theories

assuming that individuals choose from behavioral alternatives the one that gives

the highest level of satisfaction or utility (Bruch and Feinberg 2017; Kroneberg

and Kalter 2012). Applications of SCE in sociology include studies on the social

embeddedness in trust situations (Buskens and Weesie 2000), ethical consump-

tion (Andorfer and Liebe 2013), and discrimination (Beyer and Liebe 2015).

Both, SCE and conjoint experiments are also used in political science research,

for example, regarding the admission of immigrants (Bansak, Hainmueller, and

Hangartner 2016; Hainmueller and Hopkins 2015), climate agreements (Bech-

tel and Scheve 2013), and ethnic voting (Carlson 2015).

Liebe et al. 311

Most studies using SCEs focus on the main effects of choice attributes.

However, more recently, the relationship between those attributes and further

explanatory variables such as sociodemographics and attitudinal concepts

became a major concern, since it is more realistic and theoretically mean-

ingful to assume preference heterogeneity within a given population. Theo-

retically derived explanatory variables such as attitudes can be expected to

considerably increase the explanatory power of choice models. In principle,

these variables can be directly included in a choice model. Yet some authors

have questioned this approach because the integration of attitudinal questions

as error-free explanatory variables in a choice model biases model results

(Ben-Akiva, McFadden, et al. 2002, Ben-Akiva, Walker, et al. 2002). These

authors argue that it is crucial to account for the fact that attitude measures

must be understood as latent indicators of an unobserved “true” (psychic)

state. To add attitude measures directly to the models could potentially lead

to an endogeneity bias and misrepresentation of causality. Endogeneity bias

means that errors of the structural equation of indicators for attitudes might

be correlated with the error of the choice model (“[ ...] unobserved effects

that influence both a respondent’s choice and his/her responses to indicator

questions,” Daly et al. 2012:269). Misrepresentation of causality refers to the

argument that responses to indicator questions do not necessarily have a

causal relationship with behavioral choices.

In this article, we demonstrate that, based on an action-theoretic frame-

work, hybrid choice models can be a useful tool to model preference hetero-

geneity and modification in a population and hence to overcome potential

endogeneity bias and causal misrepresentation regarding attitudinal effects.

Hybrid choice models extend the specification of the traditional random

utility model by incorporating additional decision protocols and enrich the

underlying behavioral characterizations. These extensions comprise, among

others, flexible disturbances (e.g., factor analytic) to mimic more complex

error structures and to allow for the explicit modeling of latent psychological

factors such as attitudes. However, the term “hybrid choice model” is an

umbrella concept for different choice modeling techniques. In the following,

we focus on latent class structures to uncover preference heterogeneity and

segmentation as well as latent variable approaches to integrate attitudinal

effects and to investigate preference modification. We demonstrate that this

type of hybrid choice model, an integrated choice and latent variable model,

is especially valuable for sociological and other social science research.

In our empirical application, we investigate the relationship between atti-

tudes and choice behavior in an SCE study of ethical and political consump-

tion that was carried out in Germany in 2012 and investigates the preferences

312 Sociological Methods & Research 50(1)

for so-called Peace Products—goods that are jointly produced by Israeli and

Palestinian producers. Ethical and political consumption research deals with

consumer behavior that takes not only a product’s quality and price into

account but also the political, social, and environmental effects of its pro-

duction and marketing (e.g., Andorfer and Liebe 2012; Stolle, Hooghe, and

Micheletti 2005). Friedman (1996) distinguishes between “boycotts,” or

negative buying behavior, and “buycotts,” or positive buying behavior. Boy-

cotting denotes refusal to buy products and services that are associated with

negative political, social, and environmental (i.e., external) effects. Buycot-

ting refers to the deliberate purchase of products that are perceived to reduce

negative or generate positive external effects. Organic production is another

ethical product characteristic considered in our study; organic crops are

grown without pesticides and herbicides and are therefore associated with

environmental and human health benefits compared with conventionally

produced crops.

In our study, respondents had to evaluate different types of olive oils that

varied regarding production method (organic and nonorganic), origin (Italy,

Israel, Palestinian Territories, and jointly produced by Israeli and Palestinian

producers, so-called Peace Products), and price. Theoretical determinants

explaining the purchase of products with ethical attributes include pure altru-

ism; impure altruism or warm glow giving, social, and personal norms; trust;

and object-related attitudes (Liebe 2014; Stolle et al. 2005). In our case, we

concentrate specifically on relevant discriminatory attitudes toward Jews,

Arabs, and attitudes toward the Israel–Palestinian conflict, all of them can

be expected to affect stated preferences for products from Israel, Palestinian

territories, and Peace Products.

In the following, we discuss how the hybrid choice modeling framework

relates to economic and sociological theory. This is followed by a presenta-

tion of our SCE, a description of the results, and a discussion on the advan-

tages and disadvantages of hybrid choice modeling as a method to uncover

processes of decision-making, which are closely linked to action theories.

The Interplay of Attitudes, Preferences, and

Choice Behavior

We discuss the relationship between attitudes, preferences, and choice beha-

vior (decision-making) within the hybrid choice modeling framework as

pictured in Figure 1. The standard explanatory chain in social science

research holds that attitudes affect preferences, which in turn affect behavior.

However, what makes behavioral research difficult is the fact that attitudes

Liebe et al. 313

and preferences are theoretical, latent constructs that cannot be directly

observed by researchers. This has consequences for the adequate modeling

of decision and action theories in sociology and other social sciences. We

therefore first introduce the basic theoretical idea behind the SCE method, as

it was developed in economic research. Second, we discuss one specific

approach for capturing preference heterogeneity in a population (i.e., not all

individuals have the same “tastes”), an assumption that is very plausible in

most behavioral studies. Third, we specify how attitudes can be linked to

preferences and choice behavior by taking into account that they are latent

variables. It has to be stressed that in what follows, theory and statistical

modeling are discussed hand in hand because the statistical models are used

to represent the theoretical arguments. This is one strength of choice modeling

compared to other modeling approaches for testing theories in social science

research such as including theory-oriented variables in a regression model

without taking the underlying behavior model or assumptions into account.

Theory-guided Mapping of Preferences

Many theoretical explanations of sociological phenomena rest on the idea

that behavioral choices are associated with outcomes that can be expressed in

Explanatory

Variables

Latent Classes

/ Preference

Segmentation

Heterogeneity

Latent Variable

/ Attitudes

Decision

Process

Choice Indicators:

Stated

ledoMssalCtnetaL

edoMelb

air

taL

Choice Model

Indicators

Figure 1. Hybrid choice model of decision-making (simplified, Adapted from Walker

and Ben-Akiva 2001; Ben-Akiva, McFadden, et al. 2002).

314 Sociological Methods & Research 50(1)

Loading more pages...