Research Questions, Hypotheses and Variables for a Psychology Dissertation (2026): Worked Examples From Six Sub-fields

The proposal defense in a psychology program is usually lost or won on one page: the one where the research questions turn into hypotheses and the hypotheses name their variables. A committee reads that page looking for a chain with no missing link, from a problem in the literature to a purpose, to a question, to a testable prediction, to an independent variable, a dependent variable and every third variable that could carry or change the effect, each tied to a named instrument or task. This guide builds that chain step by step and then works it through six examples, one each from clinical, social, cognitive, developmental, industrial-organizational and health psychology, in the form your methods chapter will use.

Step 1: Get from the problem to a question that can be answered

A psychology research problem is a gap with a consequence: a relationship the literature has not tested in this population, a mechanism that has been assumed but not measured, or two findings that conflict. The purpose statement says what the study will do about it in one sentence (examine, test, compare). The research question then names the constructs and the population without yet committing to a direction: does social comparison on image-based platforms relate to loneliness in first-year undergraduates? The general craft of that sentence is covered in our guide to how to write a research question for a thesis; the psychology-specific rule is that every noun in the question must be a construct you can measure, which is why the question and the instrument list are written on the same afternoon.

Step 2: Write hypotheses in the forms psychology committees expect

Every quantitative research question yields at least one pair of hypotheses. The null hypothesis (H0) states no relationship or no difference; the alternative hypothesis (H1) states the relationship you predict. Two conventions matter. First, write the alternative as directional whenever the literature supports a direction (higher social comparison predicts greater loneliness), and reserve a non-directional hypothesis (social comparison and loneliness are related) for genuinely open questions; a directional hypothesis justifies a one-tailed test only if it was stated before the data were collected, which is the reason our guide to preregistering a study on OSF or AsPredicted belongs in the proposal timeline. Second, number the hypotheses to match the research questions (RQ1 yields H1a and H1b) so that the results chapter can report each in turn.

Three further forms appear in psychology dissertations and each needs its own statement: a mediation hypothesis (the effect of X on Y operates through M), a moderation hypothesis (the effect of X on Y differs across levels of W) and an interaction hypothesis in an experimental design (the effect of factor A depends on factor B). Write them as separate, numbered predictions rather than folding them into a single sentence, because each will be tested by a different model.

Step 3: Give every variable a role

Two errors recur in proposals. A variable is named without a role (loneliness is listed, but the reader cannot tell whether it is the outcome or a covariate), or a moderator and a mediator are confused, which changes the analysis entirely; our explainer on mediation and moderation analysis draws the two path diagrams side by side.

Step 4: Operationalize each construct with a named instrument or task

An operational definition says exactly how the construct becomes a number. In psychology that almost always means a validated scale or a standardized task, cited to its development paper, with the version, the number of items and the scoring rule. The instruments below are the ones that appear most often in US and UK psychology dissertations; the selection logic, permissions and reporting rules are covered in our guide to choosing a validated psychology scale for your dissertation.

• Depression: Patient Health Questionnaire-9 (Kroenke, Spitzer and Williams, 2001), nine items scored 0 to 3, total 0 to 27.
• Anxiety: Generalized Anxiety Disorder-7 (Spitzer, Kroenke, Williams and Löwe, 2006), seven items, total 0 to 21.
• Depression, anxiety and stress together: DASS-21, the short form of the Depression Anxiety Stress Scales (Lovibond and Lovibond, 1995).
• Perceived stress: Perceived Stress Scale (Cohen, Kamarck and Mermelstein, 1983), ten-item form most common.
• Loneliness: UCLA Loneliness Scale, Version 3 (Russell, 1996), twenty items.
• Affect: Positive and Negative Affect Schedule (Watson, Clark and Tellegen, 1988), two ten-item subscales.
• Personality: Big Five Inventory-2 (Soto and John, 2017), sixty items in five domains and fifteen facets, with 30-item and 15-item short forms.
• Cognitive constructs: a task rather than a scale, such as the Stroop task for inhibitory control, the n-back for working memory, or a dot-probe for attentional bias, operationalized as a reaction-time or accuracy difference score.

Step 5: Six worked examples by sub-field

Clinical psychology: a mindfulness app for anxiety

Problem: app-based mindfulness reduces anxiety in trials with general samples, but effects for students with elevated baseline anxiety are unclear. RQ: does an eight-week mindfulness app reduce anxiety in undergraduates, and does baseline severity moderate the effect? H1: participants assigned to the app will report lower GAD-7 scores at week eight than waitlist participants, adjusting for baseline. H2: the reduction will be larger for participants with moderate or severe baseline anxiety. Variables: IV, condition (app versus waitlist, randomized); DV, GAD-7 at week eight; moderator, baseline GAD-7 category; covariate, baseline score; manipulation check, minutes of app use. Analysis: ANCOVA, then a condition by baseline interaction.

Social psychology: social media and loneliness

Problem: the association between social media use and loneliness is inconsistent, and a comparison mechanism has been proposed but rarely measured. RQ: does time on image-based platforms predict loneliness in first-year students, and is the relationship mediated by social comparison? H1: daily minutes on image-based platforms will be positively associated with UCLA Loneliness Scale scores. H2: social comparison orientation will mediate the association. Variables: predictor, screen-time report from device settings; outcome, loneliness; mediator, social comparison orientation scale; covariates, gender, living arrangement. Analysis: regression with a bootstrapped indirect effect.

Cognitive psychology: working memory load and inhibitory control

Problem: models disagree on whether inhibitory control depends on available working memory. RQ: does concurrent working memory load increase Stroop interference? H1: Stroop interference (incongruent minus congruent reaction time) will be larger under high load (three-item digit maintenance) than under low load (one item). Variables: IV, load (within-subjects, two levels, counterbalanced); DV, interference score in milliseconds; covariate, none, because each participant serves as their own control; manipulation check, digit recall accuracy. Analysis: paired-samples t-test or repeated-measures ANOVA, with effect size.

Developmental psychology: parental warmth and adolescent self-esteem

Problem: cross-sectional links between warmth and self-esteem cannot establish direction. RQ: does parental warmth at age 13 predict self-esteem at age 15 after accounting for self-esteem at 13? H1: warmth at wave 1 will positively predict self-esteem at wave 2, controlling for wave 1 self-esteem. Variables: predictor, adolescent-reported warmth; outcome, self-esteem at wave 2 (Rosenberg Self-Esteem Scale); covariate, wave 1 self-esteem; moderator, adolescent gender. Analysis: cross-lagged regression using a two-wave dataset.

Industrial-organizational psychology: leadership and burnout

Problem: transformational leadership relates to lower burnout, but the mechanism is contested. RQ: is the relationship between transformational leadership and employee burnout mediated by psychological safety? H1: transformational leadership will be negatively associated with emotional exhaustion. H2: psychological safety will mediate this association. Variables: predictor, Multifactor Leadership Questionnaire ratings of the supervisor; outcome, emotional exhaustion subscale of the Maslach Burnout Inventory; mediator, psychological safety scale; covariates, tenure, hours. Analysis: mediation model; employees nested in teams, so cluster-robust or multilevel estimation.

Health psychology: self-efficacy and medication adherence

Problem: adherence interventions rarely test which belief drives behavior change. RQ: does medication self-efficacy predict adherence in adults with hypertension, and does perceived stress moderate the relationship? H1: self-efficacy will be positively associated with adherence. H2: the association will be weaker at higher levels of perceived stress. Variables: predictor, medication self-efficacy scale; outcome, adherence (self-report scale, or pharmacy refill ratio where available); moderator, Perceived Stress Scale; covariates, age, number of medications. Analysis: moderated regression with simple slopes at low and high stress.

Step 6: Align each hypothesis with a test and a sample size

Each row is a different sample size, and interaction and mediation hypotheses need far more participants than main-effect hypotheses; our guide to sample size and power analysis with G*Power walks through the inputs. If the power calculation says the moderation hypothesis needs 400 participants and you can recruit 120, the honest proposal drops H2 or labels it exploratory, rather than testing it underpowered and reporting a null.

Five failures committees flag on this page

1. A hypothesis with a construct that no listed instrument measures.
2. A directional hypothesis added after the data were seen, with a one-tailed test attached.
3. A mediator that was measured at the same moment as the outcome, described as a causal mechanism.
4. A covariate list chosen by what was convenient to collect rather than by what the literature says predicts the outcome.
5. Hypotheses that do not map one to one onto the analyses in the results chapter, which makes the discussion chapter impossible to organize. The full chapter structure and the APA 7 reporting rules are set out in our guide to how to write a psychology dissertation.

Frequently asked questions

Do I need a null hypothesis for every research question?

For every quantitative hypothesis that will be tested with inferential statistics, yes, and most programs want it written out for the proposal even if the results chapter reports only the alternative. Qualitative research questions do not carry hypotheses.

Should my hypothesis be directional?

If the literature or theory predicts a direction, state it and preregister it; if the evidence genuinely points both ways, state a non-directional hypothesis and use a two-tailed test. Never convert a non-directional hypothesis to a directional one after seeing the data.

What is the difference between a mediator and a moderator?

A mediator explains how or why the predictor affects the outcome and sits on the causal path between them; a moderator explains when or for whom the effect is stronger and changes the size of the path without being on it.

How many hypotheses should a psychology dissertation have?

As many as the research questions require and the sample can test with adequate power, which for a master’s or undergraduate project is usually two to four and for a doctoral study rarely more than six. Label any further predictions as exploratory.

Can a covariate also be a moderator?

Not in the same model with the same role. If you expect a variable to change the effect, it is a moderator and enters as an interaction; if you only want to remove its influence on the outcome, it is a covariate. State which, and why, in the proposal.

What if my instrument does not exactly match my construct?

Report the mismatch as a limitation rather than renaming the construct. Committees accept a validated scale that measures a close relative of the construct; they do not accept a construct redefined after the fact to fit the scale.