Triangulation in research: meaning and application

A survey points one way, the interviews another, and the usage figures tell a third story. Which source is right? Triangulation helps you look at the same phenomenon from several angles, so that you can draw conclusions that are better supported and more honestly bounded.

Triangulation in research means approaching the same phenomenon through several methods, data sources, researchers or theories, so that you can deliberately compare the results and support your conclusions more strongly.

When the results agree, confidence in a conclusion grows, provided the approaches do not share the same weakness. When they differ, that often produces new insight. Triangulation does not prove that something is true, and it does not turn an association into a cause.

The word comes from surveying: fixing a position from several known points makes it more precise. In research it became a metaphor: you look at one phenomenon from different angles and compare what each shows.

Two features make something triangulation. The approaches focus on the same phenomenon and the same question, and you deliberately compare the results with each other. Collecting varied data and placing it side by side in a report is not yet triangulation.

Triangulation is thus a research strategy, not a separate method, usable in qualitative, quantitative and mixed research.

The idea emerged in psychological measurement. In 1959, Campbell and Fiske argued that validation is typically convergent: a confirmation by independent measurement procedures. Validity, in their view, shows in agreement between two attempts to measure the same trait through maximally different methods. They also noted that scores partly reflect the method itself and that discriminant validation is needed too (Campbell & Fiske, 1959).

In 1966, Webb, Campbell, Schwartz and Sechrest advocated nonreactive measures: data that arise without people knowing they are being studied (Webb et al., 1966). According to a later discussion, they wanted to combine data from several classes of sources (Kelle, 2001).

The sociologist Norman Denzin turned triangulation into a fully developed research strategy in The Research Act (first published 1970). His starting point: no single method, theory or observer can capture everything that is relevant (Denzin, 2017). In 1979, Todd Jick brought the idea into organisational research and described qualitative and quantitative methods as complementary rather than as rival camps (Jick, 1979).

Denzin distinguished four types (Mathison, 1988; Carter et al., 2014):

1. Data triangulation. You use different data sources, for example data from different people, at different times or in different places.
2. Investigator triangulation. Several researchers collect or analyse the same data, so that the interpretation depends less on one person.
3. Theory triangulation. You look at the same data from different theoretical perspectives.
4. Methodological triangulation. You combine different methods, such as a survey, interviews and observation.

In practice, the types overlap. A study with a survey and interviews is methodological triangulation, but usually also data triangulation, because different people take part.

Researchers triangulate with two different intentions.

Confirming. In the original validation logic, you test a finding by checking whether information from different sources converges (Carter et al., 2014). Agreement then counts as support, contradiction as a sign of error.

Deepening. In 1992, Uwe Flick asked whether triangulation is a strategy of validation or an alternative to it (Flick, 1992). In later work, he described triangulation as an alternative to validation that increases the scope, depth and consistency of an analysis (Kelle, 2001). You then seek a more complete picture rather than confirmation.

These two logics clash. Kelle shows that converging findings add little under the deepening logic, while contradictory findings point to an error under the validation logic (Kelle, 2001). So decide your intention in advance. Mathison names three possible outcomes of triangulation: convergence, inconsistency and contradiction. All three call for an explanation by the researcher (Mathison, 1988).

It is tempting to think that three sources saying the same thing amount to the truth. Mathison called the assumption that bias will be eliminated in a multi-method design a puzzling one that frequently goes unexamined (Mathison, 1988). Fielding and Fielding warned that combining methods can even increase the chance of error, for example when researchers misinterpret similarities between incompatible data (Fielding & Fielding, 1986, cited in Kelle, 2001).

Epidemiology makes the point sharply. Lawlor, Tilling and Davey Smith describe triangulation as integrating results from different approaches that each have different, mutually unrelated sources of bias. If approaches share the same bias, a similar answer is likely to be the biased answer rather than the correct one (Lawlor et al., 2016).

In practice: ask not only whether sources agree, but whether they share a weakness. A survey and an interview are both self-reports. If both produce socially desirable answers, they confirm each other without the conclusion being right. What an association does and does not say about cause and effect is discussed in a separate article on association and causality.

Here, triangulation mainly helps to connect three kinds of information (editorial advice):

• What people say. Surveys and interviews show opinions, reasons and expectations. They are sensitive to memory, social desirability and question wording.
• What people do. Usage, purchase or visit figures and observations show behaviour. They rarely tell you why.
• How people respond spontaneously. Association tasks, eye tracking or EEG offer another perspective on responses. Read more in How do you measure associations?, What is eye tracking? and What is EEG?. These too have limits and need interpretation.

Choose approaches whose weaknesses differ. Have open answers or interviews coded independently by two researchers and discuss differences. In reports, show which sources agree, complement each other or clash. Present agreement as support within the limits of the sources, not as proof.

1. Describe the phenomenon and the question. What do you want to understand, and for which decision?
2. Choose the intention. Do you want to confirm a finding or obtain a more complete picture?
3. Choose approaches with different weaknesses. Note for each source what it measures, among whom, at what moment and where it is vulnerable.
4. Decide in advance how you will compare. What counts as agreement, as partial agreement and as contradiction?
5. Analyse each source first. Only then draw conclusions about the whole.
6. Explain differences substantively. Find out whether a difference stems from the method, the group, the moment or the phenomenon itself.
7. Report with limits. State which conclusions rest on several sources, which on one source and what remains open.

This example is fictional and intended as an illustration.

Situation. A municipal library sees few young adults in its membership records and loan figures. Before investing in new opening hours, a different offer or a campaign, the board wants to know why.

Decision question. Why do young adults stay away, and which change would appeal to them most?

Available information. Loan and visit figures by age group and time of day, plus the option of a survey, interviews and observations.

Approach. The library points four approaches at the same phenomenon. A survey among young adults in the municipality, including non-members, asks about reasons and wishes. Interviews with a few members and former members explore those reasons. The loan and visit figures show when and for what the group comes. Observations at different days and times show how the space is actually used. Two staff members code the interviews independently.

Possible interpretation. Suppose the survey and the interviews mainly mention lack of time, while the observations show that study spaces are full in the evening with young adults who borrow nothing. Then the sources complement rather than confirm each other: the loan figures may underestimate use, and the question shifts from "why don't they come?" to "what do they come for?". The survey and interviews are both self-reports and may share a bias. No results are known in this example.

Next step. The library formulates a testable expectation, for example about longer study-area opening hours, and tests it on a small scale first. Whether a change causes more visits requires a suitable design.

• Equating more sources with better conclusions. Three weak sources with the same weakness do not add up to a strong conclusion.
• Reporting only the source that fits. Leaving out contradictions loses exactly what triangulation provides.
• Comparing different phenomena. A survey about intentions and figures on actual behaviour do not measure the same thing. Say so, rather than reading them as confirmation.
• Triangulating after the fact. Unplanned sources are hard to compare cleanly.
• Applying group results to individuals. A pattern in a group does not predict what one person will do.

Triangulation lets you support a description more strongly, show the sides of a phenomenon and indicate which conclusions rest on several independent sources and which on one.

You cannot use it to prove that a conclusion is true. Triangulation also does not demonstrate cause and effect, and it does not automatically make results valid for other groups or situations. The sources in this article come from the methodology of psychology, sociology, organisational research, nursing and epidemiology. They support the principles, not a fixed procedure or effect size for your research.

Triangulation strengthens research by looking at one phenomenon from several angles. Its value lies in the comparison: agreement between sources with different weaknesses provides support, and differences point to further questions. Decide in advance what you want to achieve, choose sources deliberately and report honestly what they do and do not show together.