Professional Documents
Culture Documents
PETER G. SWANBORN
University of Amsterdam, IJsbaanpad 9, 1076 CV Amsterdam, The Netherlands
Abstract. In the social sciences, several scientific paradigms are mutually isolated owing to
their use of specific sets of methodological criteria and quality control procedures. In this
article, the central hypothesis, to be tested by conceptual analysis and logical reasoning, is
that recommended procedures for quality control in quantitative as well as qualitative
research can be derived from a common base of regulative ideas. By 'qualitative', we mean
the complex of ethnographic, anthropological, symbolic interactionist, ethnoscience and
related approaches. A second goal is to demonstrate the use of regulative ideas as a
parsimonious and fruitful base for a comparative analysis of methodological canons.
Although our focus is on the comparison of quantitative and qualitative (or: naturalistic)
research, we also pay attention to policy research as opposed to fundamental research.
1. Introduction
The next question is which criteria are applied in evaluating research results
and striving for intersubjective agreement. There exists virtually no
difference of opinion among researchers of very different background with
respect to the concept of reliability (this point of view is illustrated in
Section 3). The person of the researcher, time and circumstances of the
measurement as well as the measuring instruments are generally regarded
as irrelevant variables, leading to stochastic error and therewith lack of
reliability. Therefore, scientists usually agree that propositions about the
empirical world should be, as much as possible, independent in at least
three respects:
— researcher-independent (intersubjective agreement in a restricted sense);
— time-independent;
22 Peter G. Swanborn
— instrument-independent.
'As much as possible' is added, because absolute independence, which
would mean objectivity (in the sense that the researcher's activities can be
completely programmed for a computer) can only be obtained with respect
to a very limited subset of research activities.2
If empirical propositions are unreliable, they are not admitted to further
debate within the scientific community. So, reliability is a very central
demand to put to all kinds of research results.
Reliability controls are in principle implemented by replication, and
calculating some coefficient of agreement. The procedures. and respective
concepts, are of course well-known:
— If, with replication over researchers it shows that knowledge is
researcherindependent (e.g. by Cohen's kappa) the label inter-researcher
reliability is legitimate. Replication is a very common procedure, also in
qualitative research. In practice, however, for many researcher-roles (e.g.
interviewer, writer of the report) replication is hard to realize; reliability
can only be assumed.
— If, with replication over time (settings) the conclusions remain stable,
we use the label stability (in quantitative research stability is generally
determined by calculating some correlation coefficient).
— If, with replication over instruments, conclusions remain the same, the
label instrumental independence would be adequate. In research, this
kind of replication usually takes the form of working with a set of items
that serve as mini-instruments. If the units of observation are ranked in an
identical way by these instruments we conclude that the scale is
internally consistent or reliable (in quantitative research usually
Cronbach's alpha is calculated). If really different instruments are used
(e.g. observation and interviewing) we calculate correlation coefficients.
In naturalistic research often the label 'triangulation' is applied when
different instruments are used, but the context of this operation is
different (see below).
It is, however, well known that reliable results do not necessarily constitute
valid results. Validity means, in a very general sense, that our propositions
describe and explain the empirical world in a correct way; in a stricter
sense: that they are free from random as well as systematic errors.
A common base for quality control criteria 23
However, once again, the mixture of correspondence- and
consensustheories of truth presents itself. Intersubjective agreement with
respect to what are, and what are not, systematic errors is far more difficult
to reach than with respect to reliability. If our propositions refer to
statistical properties of operational variables and their relations, reliability
may be sufficient. But an important part of our research results is based on
argumentation and interpretation, and here the concept of validity is at
stake.
For instance, a considerable subset of research results concerns causal
interpretations of covariances. Intersubjective agreement demands at least
some explicit attempts to falsify alternative causal interpretations (by
means of research designing and/or data analysis). The relevant criterion is
called internal validity. 3 We have to remark that causal thinking is
generally absent in naturalistic research, so the necessity for controlling for
third variables is not apparent to each qualitative researcher. If, however,
the need is felt for eliminating alternative causal interpretations, verbal
argumentation and sometimes supplementary data are used.
Another important subset of research results is expressed in the language
of theoretical concepts. The problem, of course, is whether research
findings, expressed in operational language, can be generalized to relations
on a theoretical level. Here, the concept of construct validity is at stake.
Systematic errors, concerning the measurement of not-intended concepts or
the incomplete coverage of intended concepts threaten construct validity.
Such errors, that do not disappear by repeating measurements, may be
detected
denunds intersubjective
controllability (a
necessary condition)
agreementresearcher-independency
time-
valid argumentation
efficiency
usability
Scheme 1. Regulative ideas and derived quality control demands.
Whilst in policy research one regulative idea, the striving for maximizing
information content, is lost, another one comes to the fore: the pragmatic
criterion 'does it work', or in short: usability. As a necessary condition for
usability one may, of course, refer to contractual criteria; these criteria
include obligations that stem from the contract between policy agency and
researcher, such as fixed intermediate and final dates for reports; but they
also refer to a basic agreement between policy agents and researchers on the
research goals,
Within the framework of usability, Leeuw (1983) differentiates between
implementary and strategic criteria.
By implementary criteria are meant criteria concerning the quality of the
theory or model as a base for policy measures:
d. Reality content. Are the social conditions which are assumed in our
model present in the target situation? If important assumptions are not
realized, chances for success of an otherwise usable theory can be
annihilated. Does the policy system have power over organizational means?
Are other actors present (i.e. mass communication media) that exert
counterpressures? What are the costs of the program? Are there any actors
in the field who are willing to bear the responsibility for, and carry out the
new program?
References