Correlation and Experimental Methodologies
Only two of the seven methodologies in the supplied teaching source carry a quantitative label. Both promise numerical leverage over a management question, and both impose conditions that a live project environment will test hard.
The two quantitative entries in the source's methodology table
The supplied teaching source presents its research methodologies in a single table, adapted from a research methods text that is listed in the references below. Seven methodologies appear in that table, each with a short definition and a type label. Five are labelled qualitative. Correlation and experimental are the only two labelled quantitative.
That imbalance is worth noticing. If your question genuinely needs numbers, the source gives you two named routes and no others, so the definitions of those two deserve close reading. The wider set is mapped in Research Methodologies: An Overview.
The source does not expand on either methodology beyond those sentences. Everything else on this page is drawn from other parts of the same source, or is flagged as synthesis.
Correlation: a measure of association, split two ways
The definition contains two separable ideas. The first is that correlation describes a statistical measure of association between two phenomena: two things measured, and a stated relationship between the measures. Not a mechanism, not a cause, and not a story about why.
The second idea is the split: studies "may be relational or prediction". The source does not elaborate, but the distinction is consequential enough that you should decide which one you are running before you collect anything.
THE RELATIONAL / PREDICTION SPLIT
| Aspect | Relational study | Prediction study |
|---|---|---|
| Question being asked | Do these two phenomena move together, and how strongly? | Can the measure of one be used to estimate the other? |
| What the association is for | Describing and reporting a relationship that exists in the data you hold | Estimating an unobserved or future value from an observed one |
| What has to be defended | That both phenomena were measured consistently and that the association is reported as association | The same, plus the case that the relationship holds beyond the data it was derived from |
| Most common failure | Reporting the association in causal language | Extending the estimate to a population, period or context the data never covered |
The source states only that correlation studies "may be relational or prediction". The contrasts here are the implications of that split, not additional source content.
Both variants share a measurement precondition. The source describes quantitative data as information represented "using a numerical scale of some sort which is based on some agreed measurement process". If one of your two phenomena is a category or a ranked opinion rather than a measure, settle the data-type question first, using Attribute, Ordinal and Numerical Data.
Experimental: isolate, control, manipulate, observe
The source's experimental definition packs four obligations into one sentence. Separating them shows how demanding the methodology actually is.
- Identify every relevant condition in the event being investigated. You cannot control what you have not named.
- Isolate and control those conditions, so that they do not vary while you are watching.
- Manipulate the condition of interest deliberately, rather than waiting for it to vary on its own.
- Observe the effects of that manipulation, against the controlled background.
The word doing the heavy lifting is every. Not the main conditions, and not the conditions you can reach. That is a description of a setting you own, rather than one you are a participant in.
The posture is deductive: you decide what you expect, then arrange conditions that will show whether it happens. That is the logic set out in Inductive and Deductive Reasoning in Research, and it fits the positivist position the source describes as hypotheses "tested using objectively collected factual data" — see Positivism and Post-Positivism.
Why full experimental control is rare inside a live project
Read the experimental entry next to the action research entry in the same table and the source is, in effect, drawing a line between two settings.
What the experimental definition assumes
- All relevant conditions can be named in advance
- Those conditions can be held constant for the duration
- One condition can be changed deliberately, on your schedule
- The researcher has authority over the setting
What a live delivery environment supplies
- Conditions discovered part-way through, not listed up front
- Scope, staffing and priorities that move for reasons unrelated to your study
- Several things changing at once, all plausibly relevant
- A researcher who is one stakeholder among many
The consequence is straightforward. If you cannot isolate and control the relevant conditions, do not label the study experimental. Report what you had: an observed association, an uncontrolled before-and-after comparison, or a problem-solving intervention in a real setting. The last has its own entry in the source, covered in Evaluation and Action Research Methodologies.
Deciding which of the two, if either, fits
Selection rules built from the source's own definitions
None of these rules is stated in the source. Each restates a condition the source's own definitions contain, ordered so you can apply them. The broader procedure is in Choosing a Research Methodology.
Specifying a quantitative design that will survive review
Six specification steps
Fix the question and name the two phenomena
Write the question in one sentence and underline the two things it relates. If you cannot find exactly two, you are not describing a correlation study.
Define the measurement for each phenomenon
State the numerical scale and the agreed measurement process behind it. Reviewers will test the weaker of the two.
Declare relational, prediction, or manipulation
Relational and prediction are the source's split within correlation. Deliberate manipulation moves you into the experimental entry, with its control obligations.
State how cases will be selected
The source says quantitative collection relies on random sampling and structured instruments, and that where the full dataset cannot be obtained a random sample may be used to predict the population. Say which you are doing.
Name what you cannot control
List the conditions you know are moving and cannot hold still. That paragraph does more for credibility than any amount of statistical presentation.
Pre-commit to the wording of the claim
Draft the sentence you expect to write in the conclusion. If it contains a causal verb and the design has no controls, fix one or the other now.
The source does not prescribe this sequence; it is assembled from the source's own statements about quantitative data, sampling and design. The wider decision it sits inside is covered in Step 5: Designing Your Research.
Worked illustration: two questions about the same intervention
Take an organisation that has introduced a new pre-start planning routine across a portfolio of delivery projects. Two research questions are available, and they are not the same question.
One intervention, two designs
Is planning effort associated with schedule outcome?
Both phenomena are measured across the portfolio on scales the organisation already maintains. Nothing is manipulated and nothing is held constant. The output is an association, reported as an association.
Can planning effort be used to estimate schedule outcome?
Same measures, stronger commitment. You are claiming the relationship holds outside your data, which obliges you to describe the population you are estimating for.
What happens when the routine is deliberately varied?
This needs you to name the relevant conditions, hold them constant across comparison groups, and vary the routine on your schedule rather than the portfolio's. That authority rarely exists in delivery organisations.
Where these designs fail review, and how to check yours
FAILURE MODES AND THE FIX
| Failure mode | How it shows up | The fix |
|---|---|---|
| Causal language over a correlational design | The findings chapter says one phenomenon "drives" or "reduces" another | Rewrite in association language, or add the controls that would justify the stronger claim |
| An uncontrolled comparison called an experiment | The methodology chapter claims experimental design but never states what was held constant | Relabel the design and state which conditions were free to vary |
| Categories treated as measures | Ranked survey responses are averaged and reported as if from a numerical scale | Settle the data-type question first, then pick an analysis appropriate to it |
| A reachable sample described as random | Participants are whoever replied, but the write-up says random | Describe selection as it happened and state the limits on generalising |
| Prediction claimed from relational data | The study describes an association, then the conclusion estimates future values | Decide the variant at specification time, not at writing-up time |
The pattern behind all five is the same: a claim that outruns the design that produced it. The source's insistence that every step of the research fit the nature, aim, population and context of the study is the corrective, developed in Methodological Coherence: Making Every Step Fit.
Confirm before data collection starts
- The question names exactly two phenomena, and both can be measured on an agreed scale
- The measurement process for each is documented and repeatable
- You have declared in writing whether the study is relational, prediction or experimental
- If experimental: every relevant condition is listed, and you can state how each is held constant
- Case selection is described as it will actually happen, not as you wish it happened
- The conditions you cannot control are written down and will appear in the limitations
What to carry forward
- Correlation and experimental are the only two methodologies the supplied source labels quantitative; its definitions are the whole of what it gives you.
- Decide relational versus prediction at specification time. Each obliges you to defend something different.
- The experimental definition requires every relevant condition to be isolated and controlled, not just the ones within reach.
- The source's action research entry concedes that real-world work has little or no control over independent variables. Apply that concession honestly to your own setting.
- For statistical procedure, sample size and analysis technique you must go beyond the supplied source and cite the text you followed.
Frequently asked questions
What is the difference between a relational and a prediction correlation study?
The supplied source states only that correlation studies "may be relational or prediction" and does not elaborate. In practice a relational study reports how strongly two measured phenomena move together in the data you hold, while a prediction study uses that relationship to estimate one phenomenon from the other. The prediction variant carries the extra obligation of defending that the relationship holds beyond the data it was derived from.
Can I run a true experiment on a live project?
Rarely, in the sense the source's definition requires. That definition asks for every relevant condition to be isolated and controlled while one is manipulated, which assumes authority over the setting. The same source table concedes that real-world research has little or no control over independent variables, so if you cannot hold conditions constant, describe your design accurately rather than claiming experimental status.
Does a strong correlation let me claim one thing causes another?
No. The source defines correlation as describing a statistical measure of association between two phenomena, and says nothing about one producing the other. Causal claims belong to designs that control conditions. If your write-up uses causal verbs over a correlational design, either strengthen the design or rewrite the claim.
What sample size does the source recommend for a quantitative study?
None. The supplied source gives no sample size guidance, no statistical tests and no significance thresholds. It states only that quantitative collection methods rely on random sampling and structured instruments, and that where a full dataset cannot be obtained a random sample may be used to predict the population.
My data is satisfaction ratings on a five-point scale. Is that quantitative?
The supplied source treats ordered category scales of that kind as ordinal data and discusses them under its qualitative heading, while noting that counts and proportions derived from categories are often the basis for analysis. That framing is a teaching simplification and other texts classify ordinal data differently, so settle the question explicitly in your methodology chapter.
References and source attribution
- Clarke, R. J. 2005, Research methods and methodologies, pp. 35–45 — the work the supplied teaching source adapts for its table of seven research methodologies.
- Quinlan, C. 2011, Business Research Methods, 1st ed., Cengage Publishing, Chapter 1, p. 50 — the source of the methodological pyramid, on which experimental design appears as a methodology label.
- Bryman, A. 2016, Social Research Methods, 5th ed., Oxford University Press, Oxford.
- The supplied teaching source: consolidated week 2 and week 3 teaching notes and slide material on research methods, used here as the primary basis for all statements attributed to "the source".
Suggested questions for Ask KEVOS
- Rewrite my finding sentence so it reports an association rather than a cause.
- My study compares two groups but I could not control conditions — what should I call the design?
- Help me decide whether my correlation study is relational or prediction.
- List the conditions I would need to control to call this an experimental design.