Social Psychology: Research Methods

Lia Figgou, Aristotle University of Thessaloniki, Thessaloniki, Greece

Vassilis Pavlopoulos, National and Kapodistrian University of Athens, Athens, Greece
Ó 2015 Elsevier Ltd. All rights reserved.
This article is a revision of the previous edition article by C.M. Judd, volume 21, pp. 14405–14409, Ó 2001, Elsevier Ltd.

Abstract

This article presents some of the issues attached to the research methods used in Social Psychology. First it introduces the
different methods used in social psychological research (quantitative and qualitative) and discusses the basic epistemological
assumptions that lie on the basis of each of them. Then it presents methods of generating/collecting data and techniques of
quantitative and qualitative data analysis. Finally, it briefly discusses issues related to validity and reliability of qualitative and
quantitative research in Social Psychology and highlights the importance of addressing social psychological questions
through the use and analysis of empirical data.

Over the years researchers identifying their research as social Psychology, as a consequence of the development of post-
psychological have used a variety of methods to answer their modernist and post-structuralist perspectives and the so-
research questions. These can be categorized as quantitative called ‘discursive turn’ (see Social Psychological Theory,
(experimental and nonexperimental) or qualitative (analytic History of). Some basic principles of qualitative research –
techniques employed to analyze data generated through although different approaches do not share the same level
interviews and focus groups discussions, but also naturally of commitment to them – include the following: (1) a focus
occurring data). Although the division between qualitative on meaning and interpretation, (2) a preference for inductive,
and quantitative research methods coincides with debates theory-generating research, (3) sensitivity to the situated,
about Social Psychology as a discipline, here we take the context-specific nature of meaning, and (4) recognition that
stance to present the different methods in a single article, researchers’ perspectives and subjectivities are intrinsically
highlighting their epistemological differences but assuming involved into the research process (something that in quan-
that both types of methods characterize the discipline. We titative research is termed bias and it is treated as an unwel-
will briefly present the methods, their epistemological come weakness). These qualitative methods are also used in
foundations, and the questions they imply, and then we will other social science disciplines, especially in anthropology,
present some techniques of generating and analyzing data. cultural psychology, and cross-cultural psychology.
The article starts alphabetically from qualitative research
methods and then moves to quantitative research.
Methods of Collecting/Generating Qualitative Data
Interviews
Qualitative Research Methods in Social Psychology Interviewing constitutes probably the most common and
popular qualitative data collection technique. It normally
The term ‘qualitative research methods’ is employed (1) to involves a ‘dialogue’ with the researcher setting the agenda
describe ways of generating and analyzing data which are not and asking questions and the interviewee being cast in the
reducible to numbers (more often than not, texts and less role of respondent. Nevertheless, interviews as a specific type
frequently visual material) and (2) to refer to a broader of dialogue can be more or less structured. In structured
framework within which certain epistemological and interviews – rarely used in qualitative research – both the
ontological assumptions prevail (Clarke and Braun, 2013). wording and the order of the questions are the same from
While in some cases qualitative research is used as a fore- one interview to another. In unstructured interviews, on the
runner of quantitative research (analysis of interviews or other hand, a free-flowing conversational style is adopted
focus group discussions often constitutes a first step toward and respondents are encouraged to raise issues not
the construction of a survey questionnaire) or as part of a originally included in the interview schedule. Biographical
mixed-methods research design, in other cases the choice of interviews which aim at the elicitation of research partici-
qualitative methods reflects a specific way of understanding pants’ personal stories with minimum researcher prompting
social psychological phenomena and social psychological constitute a paradigmatic example of unstructured inter-
knowledge (see Social Constructionism). views. Finally, in semistructured interviews, which are most
The expansion of qualitative methods is, more often than commonly used in qualitative research, the researcher sets
not, related to the ‘crisis’ debates of the 1970s that included the agenda on the basis of their own interests and topics, but
a critique on social psychological method(s) (see Critical allows room for the participants’ more spontaneous
Psychology). However, it is not until the late 1980s and descriptions and narratives. Other distinctions are between
1990s that qualitative methods gained a foothold in Social one-to-one versus group interviewing, face-to-face versus

544 International Encyclopedia of the Social & Behavioral Sciences, 2nd edition, Volume 22 [Link]

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
 Social Psychology: Research Methods 545

telephone interviewing or interviewing through the Internet recently by social psychologists to study and understand a group
(Madill and Gough, 2008). or culture. Ethnographic observation involves participation in
a cultural-social context over a lengthy period of time.
Focus Groups
Focus groups constitute researcher/moderator-led group Structured Methods of Data Collection
discussions designed to extract opinions about a topic. They Sometimes qualitative research in Social Psychology may use
have been originally developed in market research, but they more structured methods of data collection (also often used in
gradually became a popular data generating method in quantitative research) such as open-ended questionnaires,
academic research and especially in research projects that Q-methodology and repertory grids, protocols (verbal or written
involve previously unexamined topics (Krueger and Casey, records of observations or experience, obtained in response to
2000). Focus groups provide a context which allows for the a standardized question) designed, or vignettes (a scenario is
development of argumentation and counterargumentation and provided for participants to consider and answer questions).
for the exploration of the interactional mechanisms involved in
sense making. They are also considered a method appropriate
Methods of Qualitative Data Analysis
to study groups whose voices are often marginalized within the
larger society. Qualitative Content Analysis
Interviews and focus group discussions are usually Content analysis is basically a quantitative method that
audio-recorded – researchers who intend to take into account involves establishing categories and counting the number of
nonverbal aspects of communication in their analyses tend to instances that these categories appear in a corpus of data.
video record their interviews – and then transcribed. However, the method is also employed in qualitative research
Transcription is a laborious task and demands prolonged in which systematic classification procedures are used to cate-
practice. It is also an interpretative process that requires gorize qualitative textual data into clusters of meaning (cate-
sensitivity on the part of the researcher to the nuances of oral gories or themes). Qualitative content analysis is more
speech and its differences to written language. interpretative in comparison with its quantitative counterpart
and it is interested not only in the ‘manifest’ meaning of words
Naturally Occurring Data or phrases but also in its ‘latent’ underlying meaning (Mayring,
This category includes a range of texts and interactions 2000). Coding systems consist of rules for assigning specified
produced in the course of everyday life. The researcher is units into categories. Normally, categories are intended to be
actually involved only in the sampling of the material. The unidimensional, exhaustive, and mutually exclusive, but this is
virtues of using naturally occurring data are usually highlighted not always the case in qualitative content analysis (a piece of
through their comparison to the artificiality of research inter- text may be relevant to more than one category). The imple-
views. According to critiques (Potter and Hepburn, 2005) an mentation of content analysis has been drastically affected by
interview is carried out to serve the researcher’s ends and the development of software packages, such as NVivo, Atlas ti,
agendas, which are external to the conversation itself and [Link], and the recently developed QCAmap. These
(potentially) irrelevant to the participants’ interests. Naturally tools can assist the researcher in handling and organizing large
occurring data include archival documents (ranging from quantities of data but they cannot diminish the need for
television programs and Internet materials to official/institu- intellectual effort on the part of the analyst. Content analysis
tional archival data such as health records), naturally occurring can be applied to a whole variety of data (including nonverbal
conversations (therapy sessions, telephone calls recorded in the data, such as pictures, drawings, gestures, etc.) and in relation
normal course by service providers) and – less frequently – to a variety of research questions. The fragmentation of texts
visual material such as photographs or murals. and the decontextualization of data (as instances, pieces of text
or answers are separated from their contexts) constitute
Observation potential disadvantages of the method.
It has formed the basis for much qualitative research. In common
with the category of naturally occurring data it is appropriate for Grounded Theory
the study of behavior that cannot be produced in an artificial A methodological approach was first developed by the sociol-
environment for practical or even ethical reasons. Among its ogists Glaser and Strauss (1967). Glaser and Strauss criticized
advantages is that it allows researchers to understand processes, research derived from highly abstract theories that it fails to
to understand social life as involving interrelated series of events. come up with explanations that are of relevance to those being
Different types of observation are constructed on the basis of studied. Hence, they developed a method aiming to generate
criteria such as the extent to which researchers intervene in the a local, contextual theory, grounded on data. Therefore, the
phenomenon of study or interact with research participants term grounded theory refers both to a method of inquiry, as
(Silverman, 1993). Structured observation refers to a situation well as to the theory produced by the implementation of
where the researcher creates the context where a behavior can method. The method includes an initial coding phase in which
occur. Systematic observation involves a trained researcher who researchers work systematically through the data corpus
observes and codes the phenomena of study according to a pre- generating codes (analytic labels) to describe both low-level
arranged set of criteria. Participant observation refers to a form of concepts, as well as, more abstract categories. Throughout the
systematic observation whereby the observer interacts with the process, they write memos elaborating their codes, identifying
people being observed. Ethnography is a type of observational analytic gaps, and comparing analytic categories. Constant
method used initially by cultural anthropologists and more comparison and theoretical sampling constitute principal

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
546 Social Psychology: Research Methods

analytic tasks of the method. Constant comparison involves between approaches on the ancestors and the epistemological
continually comparing elements (analytic categories and basis of the analysis, on the analytic objectives, and on the
theoretical presuppositions), while theoretical sampling conceptualization of the very notion of discourse. Potter and
involves the active sampling of new data that inform theoret- Wetherell (1987) – influenced by speech act theory, ethno-
ical categories as the analysis proceeds. The emphasis of the methodology, semiology, and post-structuralism – used the
original grounded theory method on discovering theory from term ‘discourse’ to refer to virtually any language use and
data with its empiricist connotations was criticized by scholars considered interpretative repertoires (recurrently used units of
who emphasized the constitutive nature of the analysis and content, situated in certain – usually interview – contexts and
developed more social constructionist revisions of the method oriented toward different interactional but also macro-social
(Charmaz, 2006). functions) as the units of analysis. On the other hand,
scholars influenced by Foucault’s work defined discourse as “a
Thematic Analysis system of statements which constructs an object” (Parker,
There is no agreement if it really constitutes a specific method 1990: 191) and tended to fracture texts into discrete
in its own right. It is often rather unclear which its differences discourses which subjectify speakers and reproduce power
with qualitative content analysis are, and its use has been fairly relations.
inconsistent (see, however, Braun and Clarke, 2006; for The early 1990s witnessed also the emergence of discursive
a systematic introduction to the method). Thematic analysis psychology (Edwards and Potter, 1992) which is informed by
also involves (usually inductive) coding of qualitative data into conversation analysis (CA) and prioritizes the use of natural-
clusters of similar entities, or conceptual categories and the istic data. Its focus is on the role that descriptions of the world
identification of consistent patterns and relationships between (including descriptions of psychological states) play in the
themes, so as to come up with a theoretical explanation of the management of speakers’ accountability and in the formation
phenomenon under study. of action. In the years to come, discursive psychologists
became more deeply engaged with CA, working with conver-
Interpretative Phenomenological Analysis sational corpora from everyday and institutional settings,
Its central focus is on the understanding of subjective experi- transcribed by the use of conventions developed by Gail Jef-
ence and on the meaning attached by participants themselves ferson. Between the boundary lines among this strand of
to their lived experience. Unstructured or semistructured work, which affiliates with ethnomethodological traditions,
interview data are considered to be the ideal path to the and analytic perspectives which follow Foucaultian lines,
participants’ subjectivity. In terms of initial coding and the other scholars proclaimed (Wetherell, 1998) an eclectic
process of identification of emergent themes within the corpus engagement with both CA and post-structuralism (and there-
of data Interpretative Phenomenological Analysis is also very fore engagement with both the interactional as well as the
close to some versions of thematic analysis and grounded broader ideological functions of language use) as the most
theory (Smith et al., 2009). productive starting point for discursive methods in Social
Psychology. This approach is known as critical discursive
Narrative Analysis psychology.
Narrative analysis refers to a cluster of analytic methods for
interpreting texts or visual data that have a storied form. A Conversation Analysis
common assumption of narrative methods is that people tell CA refers to a specific approach to the analysis of interaction
stories to help organize and make sense of their lives and their that emerged in the 1960s in the work of Harvey Sacks (Sacks,
storied accounts are functional, and purposeful. Different 1995). CA is interested to understand social order by focusing
approaches to narrative analysis are categorized on the basis of analytically on the sequence of talk in interaction and on the
whether they focus on the narrative content or structure, with ways participants organize mundane conversation. The recent
the thematic version interrogating what a story is about, while engagement of discursive psychologists with CA blurs the
the structural version asks how a story is composed to achieve boundaries between the two as analytic methods in Social
particular communicative aims. To this basic typology, Psychology.
according to Kohler Riessman (2008), one could also add the
dialogic/performance narrative analysis, which focusses on the Rhetorical Analysis
context and view of narratives as being multivoiced and Interest in rhetoric in Social Psychology also arose as part of the
coconstructed and the visual which links words and images in discursive turn. Key text through which rhetoric was introduced
a coherent narrative. in social psychological analyses constitutes Billig’s (1987)
‘Arguing and Thinking.’ Given his preference for scholarship,
Discursive Methods (Discourse Analysis/Discursive instead of methodology, Billig did not intend to introduce an
Psychology) analytic method in the narrow sense of the term. The text did not
There are a variety of related – but also fundamentally distinct – adhere to any specific methodology and neither included any
methods grouped under the rubric ‘discursive’ or, most methodological guidelines. Nevertheless, it served to deepen
commonly, ‘discourse analysis.’ What is common between and enrich the understanding of how to approach analytically
different methods is the recognition of the vital role of context and content in qualitative research, by advocating the
discourse in social life and an approach to language as social need to consider the rhetorical relation between topics (as units
practice, instead of a pathway to inner cognitive entities (see of analysis). According to Billig (1987), in order to capture the
Social Constructionism). Nevertheless there is little consensus meaning of any commonplace assumption, we should put it in

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
 Social Psychology: Research Methods 547

its argumentative context and examine it in relation to the effect relationships. Supporting the ontological claim of
commonplace that it aims to downgrade. Rhetorical analysis is objectivity, they strive to remain neutral and study the
also interested in relating meaning, or rather disputes over phenomena of interest ‘from a distance,’ which is expected to
meaning, to the broader historical ideological context. ensure validity of measurement and replicability of their
findings (O’Dwyer and Bernauer, 2014).
Validity and Reliability in Qualitative Research As an empirical science, Social Psychology attempts to
There is considerable debate regarding the extent to which answer questions of quantitative nature about human
validity and reliability criteria – which have been defined in behavior by testing hypotheses both in the laboratory and in
quantitative research and are imbued with its assumptions – the field. This corresponds to the distinction between exper-
are appropriate to evaluate the quality of qualitative research. imental versus nonexperimental/correlational research
According to some authors, qualitative researchers can and designs. Although sometimes presented as essentially
should incorporate criteria and techniques that deal with different or even incompatible, the difference between the
issues of reliability and validity and adjust them in their own two methods may be preferably understood in terms of the
epistemological framework. Silverman (1993), for example, level of control exerted over the variables under study (Tajfel
maintained that triangulation techniques (of researchers or/ and Fraser, 1978).
and of data sources) can vitally contribute to the quality of
research inquiry and can be compatible with the construc-
Methods of Collecting/Generating Quantitative Data
tionist epistemology that underlies many qualitative research
methods, if they are considered as highlighting the situated Experimental
use of different accounts, instead of being treated as a means Although social psychologists employ a variety of methods to
to prioritize one account vis-à-vis others. explore their research questions, the experimental paradigm
Other researchers, however, prefer to replace validity and has largely dominated the field, being characterized as the
reliability with terms and criteria more compatible to the workhorse of Social Psychology (Wilson et al., 2010). An
interpretivist background of qualitative research. Potter and experiment involves measuring the effect of different condi-
Wetherell (1987) suggested coherence of the analytic frame- tions, intentionally manipulated by the researcher, on
work (the potential of the analytic framework to give a behavioral outcome of interest. The former constitute the
coherence in a body of data by accounting both for regularity levels of an independent variable, while the latter is referred to
and variability within it), fruitfulness (the extent to which the as dependent variable. Not surprisingly, experiments are
analytic scheme generates novel explanations), new problems usually conducted in laboratory settings, which offer maximum
(the identification of contradictions and exceptions from the control over the independent variable and minimal intrusion
explanatory scheme that necessitates new analytic questions of third factors, thus allowing for testing hypotheses regarding
and answers), and participants’ own orientation (the way in causal effects.
which participants themselves see what analysts may The main advantage of an experimental design relies exactly
consider as consistent, contradictory, etc.) as validation on the ability to ensure all three preconditions for establishing
criteria of discourse analysis. Interest in understanding the causality, namely temporal precedence, covariation of the
phenomena from the participant’s view as a validation cause and effect, and exclusion of alternative plausible expla-
criterion of the quality of the analysis has been developed nations (Trochim and Donnelly, 2007). This is not a simple
also in other methods. Grounded theorists, for example, also task though: researchers must be careful to randomly assign
maintained that the extent to which results may be recog- individuals across the experimental conditions, to disentangle
nizable and may be of relevance to those studied constitutes confounded variables, to avoid extreme conditions (that would
a vital criterion of research quality. Finally, a usually lead to floor or ceiling effects), to exclude demand character-
mentioned criterion of qualitative analysis’s validation istics of experimental settings (that would lead to biased
concerns the extent to which researchers’ reflexivity is built responses of participants), and to favor experimental versus
into the analytic process. mundane realism, i.e., ‘true’ psychological impact versus
intrusive effect of everyday encounters (Hogg and Vaughan,
2010).
Quantitative Research Methods in Social Psychology On the other hand, having enough control over the
experimental conditions usually implies that the results
Quantitative research is one that relies primarily on infor- cannot be generalized to everyday life, which means that
mation of quantitative (i.e., numerical) nature. It conceptu- there is a trade-off between internal and external validity. This
alizes reality in terms of variables, which measures and can be dealt with by conducting field experiments, where
explores by applying statistical techniques in order to study experimental conditions are manipulated in real-life settings.
relationships between them (Punch, 2014). Quantitative However, some researchers warn that field experiments do
research is considered a ‘top-down’ approach in the sense that not actually increase external validity as findings may or may
specific hypotheses deriving from a theory are tested through not generalize to different real-life settings (Dipboye and
data analysis. Following the epistemological tradition of Flanagan, 1979). Beyond the issue of validity, it may not be
determinism, according to which events are accounted for by ethically acceptable or even possible for researchers to
one or more causes, it assumes that behavior is explainable. perform controlled experiments. In such occasions, nonex-
Therefore, quantitative researchers try to make probabilistic perimental methods are employed in order to collect quan-
predictions and generalizations by identifying cause-and- titative data.

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
548 Social Psychology: Research Methods

Nonexperimental dynamic, rapidly growing field of research. Of special attention

Quantitative nonexperimental methods refer to correlational are phenomena unique to the virtual environment, such as
studies, which focus on the naturally occurring associations personal Web pages, forums, and the various forms of social
among two or more factors. The use of advanced statistical networking (e.g., Facebook, Twitter, etc.). At the same time,
software gave a boost in the number of variables, as well as in a number of concerns have been raised against Internet studies,
the complexity of the relationships, that can be examined focusing on issues such as the characteristics of participants, the
simultaneously in a correlational study (see the next section). potentially fake identity of responders, or the reduced replica-
However, one should always bear in mind that any bility and generalizability of findings compared to other
correlation-based technique, no matter how sophisticated, methods. Although some of these issues have been addressed
cannot determine causal effects. To compensate for reduced (e.g., Gosling et al., 2004), the physical distance between
internal validity, the enhanced external validity associated with researcher and participant remains a major drawback and, at
nonexperimental designs contributes to the development of the same time, a distinctive advantage of Internet research.
social psychological theory by providing robustness (i.e.,
replicability of findings), representativeness (i.e., real-world Trace Measures
processes), and social relevance (Brewer, 2000). These studies are based on the fact that certain social behaviors,
The most common method of data collection in attitudes, cognitions, and emotions leave physical traces.
a nonexperimental design is a survey study. Related Although Reis and Gosling (2010) accept that the environ-
developments – most notably, the widespread use of infor- mental evidence of social psychological behaviors is largely
mation technology – have broadened the range of choice of untapped by researchers, they stress the potential value of trace
researchers to include techniques such as Internet research, measures, as the manifestations of an individual’s thoughts,
diary methods, ambulatory assessment, and trace measures (see emotions, and actions go beyond physical environments.
Reis and Gosling, 2010). These are briefly introduced below:
Methods of Quantitative Data Analysis
Survey
In quantitative studies, survey data are usually collected through Statistical methods are important tools for social psychologists
the administration of self-report scales or questionnaires using in order to explore their research questions or examine specific
the paper-and-pencil method. Surveys can obtain a large hypotheses. These tasks typically correspond to the distinction
amount of information from a large number of participants in between exploratory and confirmatory data analysis introduced
a short period of time. The psychometric properties (i.e., reli- by Tukey (1980), who wisely warned that confusing the two
ability, validity) of the measures, random sampling (in order to types of analysis can lead to systematic bias due to issues
maximize representativeness of the general population), inherent in testing hypotheses suggested by the data. However,
respecting the code of ethics (e.g., informed consent, anonymity, as Judd and Kenny (2010) point out, most data analysis in Social
and confidentiality of responses), and – of course – imple- Psychology is essentially confirmatory, in the sense that
menting the appropriate statistical techniques are some key- researchers are guided by overtly or implicitly causal theories
issues to a successful survey. The numerous types of attitudes they seek to confirm. These models often include both direct and
and values research are typical examples of survey studies. indirect effects; they may also expand to incorporate different
levels of explanation, such as the intrapersonal, the interper-
Diary Methods sonal, the positional, and the ideological (Doise, 1986). In order
Diary studies adopt a within-subject approach by collecting to understand such complex relations in their data, researchers
repeated measures from the same number of participants at are supported by recent developments in advanced statistical
different time points. They are designed to capture life in real- techniques, such as structural equation modeling (SEM) and
world settings. Their use became more popular along with the multilevel modeling (MLM), usually implemented by powerful
development of multilevel statistical software, such as Hierar- software. In the following, these developments in quantitative
chical Linear Model (HLM) (see below). Topics of interest in data analysis will be briefly presented with respect to two issues
diary studies are, among others, personality traits, emotions, of core importance in social psychological research, namely
social interaction, marital and family interaction, stress, and causal modeling and levels of analysis.
subjective well-being.
Causal Modeling
Ambulatory Assessment A causal model represents the relations between a given set of
This refers to the use of mechanical or electronic devices that variables. It can take the form of a statistical equation and it is
record information about an individual’s activity or state in usually graphically depicted. SEM is in fact an extension of
natural, everyday settings. Some applications include ambula- regression analysis for testing and estimating causal relations,
tory cardiovascular monitoring, electronic recording of the which also allows for inclusion of latent constructs represented
acoustic environment, activity monitoring, and location by a number of measured variables. The existing, constantly
mapping (Reis and Gosling, 2010). evolving SEM statistical software, such as AMOS, EQS, LISREL,
and Mplus, providing powerful controls in what concerns
Internet Research model testing, modeling error, treatment of missing data, and
Initiated in the 1990s as a convenient means to increase the testing invariance across multiple groups.
number of participants, complementary to the traditional Causal models guide social psychological research by
paper-and-pencil method, the Internet soon became a new, setting the ground for building and testing specific hypotheses.

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
 Social Psychology: Research Methods 549

Statistical handling of data collected within this framework can between two variables is affected by the presence of a third
be quite complicated as it may include multiple variables and variable, known as the moderator. Moderation suggests that
their relations are examined simultaneously in a number of the characteristics of the relation between an independent
combinations, which goes far beyond the traditional analysis factor and an outcome vary at different levels of a moderator,
of variance (ANOVA) and multiple regression techniques. Even while mediation suggests that the relation between an inde-
in simple experimental designs as well as in nonexperimental pendent factor and an outcome is fully or partially explained by
correlational studies, research hypotheses may refer to indirect a mediator. Statistically speaking, moderation represents an
effects, which involve the inclusion of third variables explain- interaction or a product in factorial ANOVA or in multiple
ing or modifying the relations between an independent factor regression, depending on whether the moderator is qualita-
and an outcome. Baron and Kenny’s (1986) seminal work tively or quantitatively measured, respectively (see Hayes and
discussed in depth the conceptual, strategic, and statistical Matthes, 2009; for computational procedures of calculating
considerations of distinguishing between two types of third interaction terms). From a conceptual point of view, hypoth-
factors affecting a causal relation, i.e., the mediator and eses involving moderation address the question of invariance
moderator variables. of causal relations between an independent variable and an
outcome across different units, such as persons, situations, or
Mediation cultures. In fact, any causal assumption implies some degree of
A simple mediation model suggests that the observed relation invariance across persons. Judd and Kenny (2010) remind us of
between an independent variable and an outcome can be other sources of invariance including time points, situations,
explained by the effect of a third factor, known as mediator. experimental stimuli (e.g., evaluative priming procedures),
According to Baron and Kenny (1986), three prerequisites are specific statistical techniques (e.g., meta-analysis, logistic
necessary in order to establish mediation: (1) the independent regression), and cross-cultural comparisons. Invariance indi-
variable should be a significant predictor of the dependent cates the replicability of our findings and the limitations in
variable; (2) similarly, the mediator should have a significant generalizing our conclusions. In methodological terms,
effect on the dependent variable; and (3) the mediator should moderation analysis provides evidence of external validity as it
be a significant predictor of the dependent variable while examines to what extent a causal effect is considered universal.
controlling for the effect of the independent variable. A relation Statistical handling of data in moderation analyses goes
is fully mediated when the direct path from the independent beyond simple identification of a significant partial effect of
variable to the outcome becomes nonsignificant after a product predictor in a linear model. Common problems
accounting for the effect of the mediator. Partial mediation include, among others, multicollinearity and power. Multi-
occurs when the mediator accounts for some, but not all, of the collinearity may yield coefficients with higher standard errors
relation between the independent variable and the outcome. since the interaction term of two continuous variables (i.e., the
The amount of mediation is called the indirect effect. moderator) inevitably correlates highly with the two main
A simple, common technique to identify mediation is effects used to calculate it. Low statistical power indicates that
Sobel’s test (1982), which compares the difference in the replication of a significant moderation effect is difficult to
relation between the independent variable and the outcome achieve, especially with continuously measured independent
before and after inclusion of the mediator in the regression variables, and calls for the use of large sample sizes.
equation. More recently, Preacher and Hayes’s (2008) In addition to the assumptions for applying specific statis-
nonparametric bootstrap method has become increasingly tical procedures, Judd and Kenny (2010) elaborate on some
popular, which is recommended for small samples as it does more perplexing issues of moderation testing. Not surprisingly,
not violate assumptions of normality. SEM programs or espe- their discussion raises the problem of causality from a theoret-
cially written statistical software macros can be applied in order ical perspective once again. They show that the direction of
to test for complex cases of mediation, such as multiple inde- causality between A and B (i.e., the decision on which variable
pendent factors or outcomes, multiple mediators or latent will be treated as independent or outcome) matters a great deal
variables used as mediators (see, for example, the Web pages of in moderation testing, especially when the homogeneity of
Andrew Hayes, [Link] and Kristopher variance assumption is violated, so that A or B vary differently
Preacher, [Link] at high and low levels of the moderator. In such statistically
Beyond statistical complexity, it should be underlined that ambiguous situations, a solid causal theory is necessary to
mediation is primarily a conceptual issue, i.e., the conclusions indicate how moderation is assessed.
from a mediation analysis are valid only if the causal
assumptions hold true (Judd and Kenny, 2010). Therefore, Mediated Moderation and Moderated Mediation
researchers should make sure that reverse causal effects (e.g., It is not rare that both mediator and moderator variables are
the mediator being caused by the outcome) or confounding integrated in a causal model. In these cases different hypotheses
variables (i.e., factors causing both the mediator and the can be drawn, which are extensions of the simple mediation
dependent variable) are not present. Also, the mediator should and moderation frameworks and they are defined accordingly.
be reliably measured in order to minimize bias. In moderated mediation, the mediating effect of a variable in the
relation between an independent factor and an outcome varies
Moderation across different levels of a moderator (Preacher et al., 2007). In
If mediation analysis attempts to explain ‘how’ or ‘why’ an mediated moderation, the effect of a variable moderating the
effect occurs, moderation refers to the question of ‘when’ this relation between an independent factor and an outcome is
effect holds, i.e., if the strength and/or direction of the relation accounted by the mediating process of another variable (Muller

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
550 Social Psychology: Research Methods

et al., 2005). Therefore, moderated mediation and mediated Additionally, in social psychological research inferences are
moderation do not necessarily imply the existence of two made for a unit of analysis by studying random samples. MLM
different data sets, but rather two different causal models takes into account simultaneously the sampling error at
guiding research hypotheses, which can be based on the same different levels, which is not the case with traditional OLS.
data set. Therefore, MLM produces more accurate estimates than OLS
Although mediation and moderation analyses are quite because it considers the reliability of scores and the differences
common in social psychological research literature, mediated in sample sizes. These advantages are pronounced when
moderation and moderated mediation hypotheses are not hypotheses of interest concern within-unit relations, and when
frequently examined, which would be particularly informative the data structure is irregular, for example, due to missing data
in extending the theoretical models used. This may be due to (Nezlek, 2008).
the statistical complexity of such questions, as computational MLM is appropriate when research questions involve more
routines to conduct these analyses are only recently being proximal (L1) and more distal (L2, L3.) factors influencing
developed. Fairchild and MacKinnon (2009) provide illus- individual behavior, or simply when the data are hierarchically
trated examples of the equations required to simultaneously structured. Two issues of relevance need to be considered here:
estimate mediation and moderation effects; they also underline (1) the nature of the variables used; and (2) the nature of the
the contribution of these techniques in applied settings, where relations between levels. Variables in a multilevel model may
the questions of ‘how’ and ‘for whom’ an intervention program be intrinsic or derived (either aggregated or disaggregated; see
is effective become crucial. Van de Vijver and Poortinga, 2002). The latter case raises
Modern SEM computer programs, with their user-friendly concerns about equivalence, i.e., to what extent a construct
graphical interface and huge estimating capabilities, facilitate measured in different groups has the same underlying struc-
the conceptual and statistical handling of mediation and ture. Structural equivalence is a precondition in order to
moderation analyses, provided that researchers are respectful perform group mean comparisons. It can be examined by
with regard to consideration of assumptions (see Bentler and calculating a congruence coefficient (e.g., Tucker’s phi) through
Chou, 1987; for an excellent overview). Testing of alternative exploratory multilevel factor analysis procedures, or by
models on the grounds of goodness-of-fit indices is particularly applying multilevel confirmatory factor analysis with the use of
enlightening in terms of theory building. For example, longi- SEM programs. Another question of interest is to what extent
tudinal designs may benefit from examining four competing the same pattern of predictor–outcome relations holds across
hypotheses referring to stability, causality, reversed causality, L2 units, e.g., across groups, cultures, or time points. This is
and reciprocal causality, respectively. Yet, it should be typically referred to as a cross-level interaction. Conceptually
reminded that SEM is a confirmatory – as opposed to explor- similar to moderation, it is substantive for social psychologists
atory – technique, which means that it can be used to compare who seek to explore the interplay between different levels of
multiple theories that are specified a priori. explanation (Doise, 1986) and can be examined through
specialized statistical software (HLM).
Levels of Analysis
Although many social psychological theories focus on indi- Some Additional Comments
vidual processes, they are inherently placed at the interface Error
between personal and group phenomena; therefore, hypoth- Dealing with error in quantitative data analysis is almost as
eses drawn on these theories may include multiple levels of important as model testing. Although relevant assumptions
analysis. The term MLM – which is a shortcut for multilevel (such as lack of measurement error and homogeneity of error
random coefficient modeling, also known as hierarchical linear variance) are embedded in OLS analyses, researchers some-
modeling, among others – is used to describe hierarchically times tend to overlook them. Violation of these assumptions
structured data, i.e., observations at one level which are nested may lead to unfortunate conditions, such as measurement bias,
within observations at another level. For example, employees attenuation of measures of association, reduced power in
(L1) are nested within departments (L2) and organizations testing interactive effects, and underestimation of mediation
(L3); students (L1) are enrolled to different classes (L2) and effects, among others (Baron and Kenny, 1986). Therefore,
schools (L3); in cross-cultural studies, individuals (L1) come obtaining reliable measures, identifying outliers, testing for
from different countries/cultures (L2); in diary studies, nonnormality, and treating sources of error at different levels
emotions at different events or time intervals (L1) are nested are necessary steps to be taken prior to any analysis involving
within persons (L2). As Kreft and de Leeuw (1998) put it, once specific hypothesis testing.
we realize hierarchies exist, we recognize them everywhere!
In statistical terms, the key question is whether observations Variable Coding
are independent or not. In hierarchically structured data In a regression equation, the parameter estimate of a predictor
observations at L1 are not independent, which violates indicates the effect of the predictor on the criterion variable
a fundamental assumption of traditional ordinary least squares when all other predictors equal zero. Therefore, all simple effects
(OLS) techniques, such as ANOVA and regression. Single-level are meaningless if zero is not a meaningful value for the
analyses that ignore the hierarchical structure of data may predictors included in a regression model. This straightforward
provide misleading results because relations at different levels statement is not always fully understood by researchers, thus
of analyses are in fact independent. Van de Vijver and Poortinga leading to a series of misinterpretations. When a categorical
(2002) provide a taxonomy of multilevel fallacies occurring predictor is coded as 1 versus 2, then the slope of a continuous
when different patterns of relations across levels are ignored. independent variable points to a nonexistent level of the

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
 Social Psychology: Research Methods 551

categorical predictor. An easy solution for this problem is to use psychological knowledge. We devoted space both to qualitative
a coding of 0 versus 1. Even in that case, the slope of a contin- and quantitative research and we attempted to highlight not
uous independent variable is not a ‘main effect,’ as presented by only the differences between them but also the diversity within
many, but rather the simple effect of that independent variable each research tradition. Of course many supporters of each
when the categorical predictor equals zero. tradition would disagree about the extent to which a pluralistic
In the same underlying rationale, when calculating the stance toward method is preferable or even acceptable,
product for a regression interaction term it is advised that the although a mixed-methods research approach combining the
two component variables are centered prior to computing their quantitative and qualitative typologies is gaining ground in
product, so that zero is an interpretable value for both recent years (e.g., Bryman, 2006). In any case, it is hard to
predictors (Aiken and West, 1991). Of course the above is not disagree that the diverse ways of generating and analyzing
necessary if zero is already a meaningful value, like for example, empirical data and the importance attributed to this process
in Likert scales coded from
3 to þ3, instead of 1–7. In has given Social Psychology as a discipline a distinctive identity
multilevel models, variable centering can affect significance among the social sciences.
tests, but most dramatically it changes intercepts and error
terms (Kreft and de Leeuw, 1998). HLM software provides two
centering options (in addition to the uncentered solution), See also: Anthropology: Overview; Attitude Measurement;
namely grand mean centering and group mean centering, Critical Psychology; Cross-Cultural Psychology; Cultural
depending on whether the slopes represent deviations from the Psychology; Implicit Social Cognition; Indigenous Psychology;
overall mean or from each group’s mean, respectively. With Levels of Analysis in Social Psychology; Social
grand mean centering, estimates of slopes include between Constructionism; Social Psychological Theory, History of;
group differences in means of predictors, which are not Social Psychology.
included in group mean centering. At L2 (in two-level models),
it is preferred to grand mean center, as this helps interpreting
the intercept. At L1, group mean centering is closest to con- Bibliography
ducting within group regression analysis.
Aiken, L.S., West, S.G., 1991. Multiple Regression: Testing and Interpreting
Interactions. Sage, Newbury Park, CA.
Missing Data Baron, R.M., Kenny, D.A., 1986. The moderator-mediator variable distinction in social
Missing data are quite underestimated by researchers. In any psychological research: conceptual, strategic, and statistical considerations.
paper describing a quantitative statistical method there is Journal of Personality and Social Psychology 51, 1173–1182.
a section devoted in how to treat missing values, often Bentler, P.M., Chou, C.-P., 1987. Practical issues in structural modeling. Sociological
Methods & Research 16, 78–117.
skipped by some readers. However, the presence of missing Bryman, A., 2006. Integrating quantitative and qualitative research: how is it done?
observations may have important implications, from Qualitative Research 6 (1), 97–113.
restricting the generalizability of the findings to practically Billig, M., 1987. Arguing and Thinking: A Rhetorical Approach to Social Psychology.
preventing an analysis from running at all. A typology Cambridge University Press, Cambridge, UK.
Brewer, M.B., 2000. Research design and issues of validity. In: Reis, H.T., Judd, C.
describing missing data in terms of their underlying cause is
(Eds.), Handbook of Research Methods in Social Psychology. Cambridge University
owed to Rubin (1987): missing completely at random, Press, New York, pp. 3–16.
missing at random, and missing not at random. The strategies Braun, V., Clarke, V., 2006. Using thematic analysis in psychology. Qualitative
for handling missing data include case (listwise) deletion, Research in Psychology 3 (2), 77–101.
imputation by using a substitute value (such as the group Charmaz, K., 2006. Constructing Grounded Theory: A Practical Guide through
Qualitative Analysis. Sage, London.
mean or a multiple regression estimate), and multiple Clarke, V., Braun, V., 2013. Successful Qualitative Research: A Practical Guide for
imputation (a method that estimates the sampling variance Beginners. Sage, London.
due to imputation). These strategies differ in terms of Dipboye, R., Flanagan, M.F., 1979. Research settings in industrial and organizational
sophistication, which in turn attempts to compensate for psychology: are findings in the field more generalizable than the laboratory?
American Psychologist 34 (2), 141–150.
their disadvantages. Therefore, they should be studied care-
Doise, W., 1986. Levels of Explanation in Social Psychology. Cambridge University
fully before being applied in a given data set through modern Press, Cambridge, UK.
powerful statistical software. Edwards, D., Potter, J., 1992. Discursive Psychology. Sage, London.
Judd and Kenny (2010) urge social psychologists to be Fairchild, A.J., MacKinnon, D.P., 2009. A general model for testing mediation and
more creative in how they deal with missing data, which they moderation effects. Preventive Science 10 (2), 87–99.
Glaser, B.G., Strauss, A., 1967. The Discovery of Grounded Theory: Strategies for
believe can lead to new insights about their research designs. Qualitative Research. Aldine Publishing Co, Chicago, IL.
Overall, perceiving statistical analyses as a challenging oppor- Gosling, S.D., Vazire, S., Srivastava, S., John, O.P., 2004. Should we trust web-based
tunity to discover rather than a routine of imposed restrictions studies? A comparative analysis of six preconceptions about internet
will help researchers get the most out of their data and improve questionnaires. American Psychologist 59, 93–104.
Hayes, A.F., Matthes, J., 2009. Computational procedures for probing interactions in
their theories.
OLS and logistic regression: SPSS and SAS implementations. Behavior Research
Methods 41, 924–936.
Hogg, M., Vaughan, G., 2010. Social Psychology, sixth ed. Prentice Hall, London.
Conclusion Judd, C.M., Kenny, D.A., 2010. Data analysis in social psychology: recent and
recurring issues. In: Fiske, S.T., Gilbert, D.T., Lindzey, G. (Eds.), Handbook of
Social Psychology, fifth ed. Wiley, Hoboken, NJ, pp. 115–139.
In this article, we presented the methods used by social Kohler Riessman, C., 2008. Narrative Methods for the Human Sciences. Sage,
psychologists in order to set and address research questions, to Thousand Oaks, CA.
generate and analyze empirical data, and to contribute to social Kreft, I.G.G., de Leeuw, J., 1998. Introducing Multilevel Modeling. Sage, Newbury Park, CA.

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552
552 Social Psychology: Research Methods

Krueger, R.A., Casey, M.A., 2000. Focus Groups: A Practical Guide for Applied Reis, H.T., Gosling, S.D., 2010. Social psychological methods outside the laboratory.
Research, third ed. Sage, Thousand Oaks, CA. In: Fiske, S.T., Gilbert, D.T., Lindzey, G. (Eds.), Handbook of Social Psychology,
Madill, A., Gough, B., 2008. Qualitative research and its place in psychological fifth ed. Wiley, Hoboken, NJ, pp. 82–114.
science. Psychological Methods 13 (3), 254–271. Rubin, D.B., 1987. Multiple Imputation for Nonresponse in Surveys. Wiley, New York.
Mayring, P., 2000. Qualitative content analysis. Forum: Qualitative Social Research 1 (2). Sacks, H., 1995. Lectures on Conversation. Blackwell Publishing, Oxford, UK.
Retrieved from: [Link] Silverman, D., 1993. Interpreting Qualitative Data: Methods for Analysing Talk,
Muller, D., Judd, C.M., Yzerbyt, V.Y., 2005. When moderation is mediated and mediation Text and Interaction. Sage, London.
is moderated. Journal of Personality and Social Psychology 89, 852–863. Smith, J.A., Flowers, P., Larkin, M., 2009. Interpretative Phenomenological Analysis:
Nezlek, J., 2008. An introduction to multilevel modeling for social and personality Theory, Method and Research. Sage, London.
psychology. Social and Personality Psychology Compass 2, 842–860. Sobel, M.E., 1982. Asymptotic confidence intervals for indirect effects in structural
O’Dwyer, L.M., Bernauer, J.A., 2014. Quantitative Research for the Qualitative equation models. Sociological Methodology 13, 290–312.
Researcher. Sage, Tousand Oaks, CA. Tajfel, H., Fraser, C., 1978. Social Psychology as social science. In: Tajfel, H., Fraser, C.
Parker, I., 1990. Discourse: definitions and contradictions. Philosophical Psychology (Eds.), Introducing Social Psychology. Penguin, Harmondsworth, UK, pp. 21–53.
3 (2), 189–203. Trochim, W., Donnelly, J.P., 2007. The Research Methods Knowledge Base, third ed.
Potter, J., Hepburn, A., 2005. Qualitative interviews in psychology: problems and Thomson, Cincinnati, OH.
possibilities. Qualitative Research in Psychology 2 (4), 281–307. Tukey, J.W., 1980. We need both exploratory and confirmatory. American Statistician
Potter, J., Wetherell, M., 1987. Discourse and Social Psychology: Beyond Attitudes 34, 23–25.
and Behaviour. Sage, London. Van de Vijver, F.J.R., Poortinga, Y.H., 2002. Structural equivalence in multilevel
Preacher, K.J., Hayes, A.F., 2008. Asymptotic and resampling strategies for assessing research. Journal of Cross-Cultural Psychology 33, 141–156.
and comparing indirect effects in multiple mediator models. Behavior Research Wetherell, M., 1998. Positioning and interpretative repertoires: conversation analysis
Methods 40, 879–891. and post-structuralism in dialogue. Discourse and Society 9 (3), 387–412.
Preacher, K.J., Rucker, D.D., Hayes, A.F., 2007. Addressing moderated mediation Wilson, T.D., Aronson, E., Carlsmith, K., 2010. The art of laboratory experimentation.
hypotheses: theory, methods, and prescriptions. Multivariate Behavioral Research In: Fiske, S.T., Gilbert, D.T., Lindzey, G. (Eds.), Handbook of Social Psychology,
42, 185–227. fifth ed. Wiley, Hoboken, NJ, pp. 51–81.
Punch, K., 2014. Introduction to Social Research: Quantitative and Qualitative
Approaches, third ed. Sage, London.

International Encyclopedia of the Social & Behavioral Sciences, Second Edition, 2015, 544–552