The WiW methodological paradigm for HRM
research
The results of
HRM research do not lead to the construction of immutable laws, but only remain
socially, culturally, and historically limited generalisations[1]. The formulation of a research program requires not only
determining the area of research, but also specifying the problem itself and
the purpose of this research[2].
What research instruments one will use in their case will be determined by the
research objective and its feasibility.
We study what is
observable, measurable, and susceptible to experimentation. Science is based on
empirical evidence.
Key terms
All data
obtained by asking employees questions are called survey data. All
participants, regardless of whether they took part in surveys, experiments, or
interviews, are called respondents, because the object of analysis is
their reactions (answers).
Results of
measuring people can have the form of numbers, in which case we speak of
quantitative research/analysis, or words, which are most often a component of
qualitative research/analysis.
Quantitative
data are sets of numbers that are subjected to statistical analysis.
Qualitative data are sets of words that are an attempt to describe different
visions of the researched phenomenon (reality is in the eye of the beholder),
subjected to the researcher's interpretative analysis, which may include objectivising
elements such as classification of statements by independent judges, counting
the frequency of using different phrases.
Quantitative
research differs from qualitative research in the degree of proceduralization
of methods of analysis. The aim of quantitative research is most often the
objective testing of hypotheses assuming relations between variables. The aim
of qualitative research is most often to identify individual ways of perceiving
reality.
Methodological
pluralism/eclecticism & pragmatism in the choice of problem
The WiW paradigm
rejects both anarchism (accepting arbitrary methods and techniques drawn
even from individual experience) and methodological fundamentalism, in
which different research methods cannot be mixed. It agrees with the postulate
that research methods in HRM should be applied reflexively, as they are
heuristic in nature, making algorithmizing impossible. Therefore, it recommends
pluralism and even methodological eclecticism that accepts the
use of methods drawn from different disciplines and theoretical approaches to
solve a research problem[3].
At the stage of
selecting the research problem, it is recommended to apply a pragmatic
approach, if the analysed research problem does not have important practical
consequences, then it is not worth dealing with it, leaving such considerations
to basic sciences.
Specificity of the test
object
Methodologists
forget that the study of inanimate objects is governed by different laws than
the study of people. To make matters worse, we are dealing with conducting
„people-by-people” research. The specificity of HRM research lies in the fact
that the objects of measurement are people who create meanings, i.e.,
their reactions to stimuli are mediated by their expectations, interpretations
determined to a large extent by the record of their previous experiences.
Therefore, in contrast to the sciences, in HRM each replication of the study
is a success, because the group of surveyed employees, their experience,
the cultural context is always changing...
The objects of
analysis in HRM research are mental facts, i.e., most often people's
answers (verbal or categorized on numerical scales) to the questions asked. It
should be remembered that this type of quantitative data is almost always
distorted, as has been shown in many studies[4].
The model of the question-answer process shows why there is such a great
variation in the responses of the respondents.
Answering a
question about evaluation, e.g., job satisfaction, requires the activation of
various information contained in long-term memory in its semantic (e.g., what
it means to be satisfied) and episodic parts (e.g., recalling various emotional
states). The recalled information, according to a concept of consciousness
called a multiple sketch model, is subject to continuous editing. At no
point in this process can it be said that the editing is complete, and the
final outcome is consciously experienced. At a given moment, we recall the
worst episodes; in an hour, we may recall information that radically changes
our judgment. When we are in a good mood, we look for positive aspects of
working in this company; when we are in a bad mood we "look for holes in
the whole". Respondents, while filling in the questionnaire, very rarely
have ready marks of satisfaction "in their heads". The assumption
that we constantly archive different opinions is not very convincing. An
alternative assumption is that we construct them on an ongoing basis when they
are needed. Specific goals, standards, judgments, and attitudes with a high
capacity to generate further information. We have various general opinions,
goals, standards, and attitudes encoded in our minds to generate further
opinions. These are essential for the formation of emotions, because without
them it is impossible to give any meaning to the events we encounter. Most of
the cognitive representations (e.g., views about the role of work in life) that
we ask about are not represented in the mind before the evaluation is
initiated. Such representations can be described as virtual (because they do
not exist before the question is asked). Our approach differs significantly
from the traditional approach of measurement theory, which assumes that the
respondent already has a fixed 'true' answer - one they would give themselves,
so the primary concern is to minimize measurement error caused by the form of
the question, the social context. Every evaluation requires the ability to
focus one's attention to select information, to omit or at least block out
those that are of peripheral importance. In the process of transforming a
thought into an utterance, a chain of associations emerges in the mind. Each
word, especially an ambiguous one, triggers a sequence of associations that run
often in different, even very divergent directions. There are many cognitive
schemas encoded in long-term memory that are "ready" to interpret
such a word. The mind usually sifts through associations and selects only those
that are related to the thought we want to express. The more accurate this information
sifting, the more effective the next stage of processing associated with
conscious attention can be. Only a modest fraction of this process can be made
conscious, but this does not mean that we cannot take control and turn our
attention to different aspects of the issue. In this way, awareness modifies
the operation of the filter. We can call up information from long-term memory,
and it will filter the incoming information. To sum up, we must be aware that respondents
very often do not have a ready answer and they form it only when the questions
are asked. Very often, they do not reproduce their opinions but construct
them. What opinion they form depends on which of the four strategies of forming
an opinion we apply: 1) reproducing ready-made judgements, 2) motivated
processing, 3) heuristic (simplified) processing, and 4) analytical (detailed)
processing.
The information
processing strategy chosen is determined by the respondent's cognitive
abilities (e.g., level of reflexivity), state of the organism (overload, mood),
and goals determining the degree of involvement. The choice is also influenced
by the characteristics of the object of assessment (degree of familiarity and
complexity) and the characteristics of the situation (time pressure, social approval,
how costly mistakes are). In surveys, respondents, due to time constraints and
the lack of costs of making an incorrect judgment, extremely rarely use an
analytical strategy. Therefore, we should keep in mind:
(1)
Importance of psychological
realism of the research - it is very important to maintain the respondents’
engagement e.g., by offering personalized feedback if it is possible. The
respondent wants to understand not only WHAT is being asked about, but also
WHY?
(2)
Respondents do not have ready
answers in their heads and must have the right to say, “I don't know”, not
applicable, or omit the answer. Forcing them to give an answer can lead to
irritation and giving random answers to subsequent questions.
(3)
Respondents, if they can, will
avoid the mental effort – they love to use middle options on the rating scale,
so even-numbered points with Don’t Know (Difficult to Say) option outside the
rating scale is recommended. Research[5]
has shown that the absence of a middle option does not significantly increase
the number of Don’t Know (Difficult to Say) answers.
Conclusion: Respondents’
answers have different validity and reliability. Sophisticated methods of data
analysis are of no use if these data are distorted in various ways.
Scientific concepts and operational
definitions
In science, we
use the language of observation and the language of theory in parallel. In the
language of theory, we use scientific concepts (theoretical constructs,
latent variables) e.g. leadership style, need for dominance, emotional
well-being of an employee etc., which have to be translated into the language
of observation.
The WiW paradigm
recognizes that the theoretical constructs under study are natural concepts
that cannot be defined in a classical way by means of necessary and sufficient
conditions, so the solution to the problem is operationism[6], which assumes
that scientific concepts do not capture the essence of things, but only
give the scientist’s actions, his psychophysical operations needed to define
the thing under study.
We use various
measurement tools to build indicators. An example would be sets of questions
built to measure an employee characteristic. Such sets of questions are called
scales (e.g., Anxiety Scale) or psychological tests, which can be treated as a
variety of calibrated tools[7].
The positivist
approach[8] to
quantitative research analysis assumes that the objects of research are facts,
which are presented in the language of variable values. Hundreds of variables
and their operationalization have been described in scientific HRM studies. One
can get the impression that the introduction of another scientific concept to
describe a person is overly accepted. That is why the researcher must choose
the variables that are the subject of his inquiries by describing the
theoretical model of the phenomenon described and the measurement model of the
theoretical constructs.
The task of the
researcher is not limited to registering facts and laws governing the facts but
consists in such an ordering of them in theoretical models as to be able to
predict subsequent facts on their basis.
Theoretical Models
In HRM,
cognition is achieved mainly through model testing rather than observation[9]. Therefore, the first step is to select, based on a literature
review, the theoretical variables (scientific concepts) that will be used to
model the phenomenon of interest to the researcher.
A theoretical
model should be as follows:
- simple - the fact that reality is complex does not imply that the
model should be complex[10],
- congruent with available scientific facts if it is not intended to
question interpretation of them,
- logical, internally consistent[11],
- able to generate predictions,
- empirically verifiable.
A theoretical
model that has been confirmed by many studies can be called a theory.
Each model in
HRM consists of an a priori part, an assumption that the selected variables are
valid and relevant, or a set of hypothetical relationships between variables,
which are subjected to precise empirical tests. In addition to the theoretical
model, a measurement model must be specified, that is, a way of
operationalizing all the variables.
Hypotheses are
falsifiable statements about the relationships between the variables specified
in the theoretical model.
Five types of triangulation
The WiW paradigm
recommends 5 types of triangulation: (1) methods, (2) data, (3)
operationalization, (4) modes of analysis, and (5) researcher.
Triangulation of methods
Even in online
surveys, we can combine correlational, experimental, and qualitative methods.
We analyse numerical answers to closed questions with quantitative methods, and
verbal answers to open questions with qualitative methods.
Data triangulation
The availability
of population representative random samples is very limited in the social
sciences, since people can be drawn but cannot be forced to participate in
surveys. Therefore, in most cases, surveys are conducted on convenience samples
consisting of people who have agreed to participate in the survey. We increase
external validity by replicating studies in different convenience samples. This
means that we should test the same hypotheses on different data sets.
Triangulation of
operationalizations
There are no
standard operationalizations of variables in HRM. Operationalization of
variables should be carefully selected considering the specifics of the sample,
e.g., the item "I make decisions under time pressure more easily" is
a good indicator of low reactivity in the group of young employees, but not
among managers. Even if we use standardized ready-made measurement tools, their
psychometric properties should be checked on the sample.
Triangulation
of analytical methods
Although in quantitative analyses assumptions are made about the
axiological neutrality of science and the non-interference of the researcher,
even in the pre-proceduralised, objectified statistical analyses, the
researcher has to make decisions about how to "clean" the data set,
how to build indicators, how to choose assumptions about the level of
measurement, how to choose statistical tests. The decision of whether to treat
questionnaire score as a continuous or ordinal variable (e.g., after median
splitting) may lead to different conclusions. Therefore, the WiW paradigm
recommend using different ways of data analysis like parametric vs
nonparametric tests on the same data, with the increased trust to the results
that are robust to the change of statistical tests.
Researcher triangulation
When analysing
qualitative data, words, researcher triangulation is recommended, data
should be coded by at least two people independently of each other.
External and internal
validity of research
We increase
external validity by using different types of triangulation – in particular, by
testing the same hypotheses on different data sets.
Where possible,
we should take care to ensure the INTERNAL VALIDITY of the study. Even in
surveys we can manipulate the independent variables - that is, we can
conduct experimental research by assigning volunteers randomly to different
experimental conditions.
Where possible,
in both surveys and interviews, we introduce DESCRIPTIONS of the objects whose
evaluation we want to know. For example, when asking employees for their
opinions about their boss, we are not able to determine to what extent it
results from the employee's perception and to what extent from the objective
characteristics of the boss. Asking for the evaluation of the model description
of e.g., a dominant, partner-like boss we can investigate individual
differences in the evaluation of various features that were the basis for the
construction of these descriptions.
Quality of Data
Before analysis,
data sets should be carefully cleaned of "false" respondents, who,
e.g., gave random answers[12].
Standard measurement tools used in research should be checked for psychometric
properties/adapted to the group of respondents studied.
Quantitative, experimental
case studies[13]
Findings on relationships between 2-3 variables (ceteris paribus) are difficult to apply in practice because of multidimensionality of reality). Therefore, WiW methodological paradigm
promotes QUANTITATIVE experimental
case studies, where the values of variables at selected time points are manipulated and quantitative measurements
are made over a long period of time.
