Research Methods & Maths Skills

Research methods and mathematical skills form the backbone of Paper 3 (Psychological Skills), where they are examined most heavily, and they are also embedded in the classic and contemporary studies you meet across Papers 1 and 2. On top of this, at least 10% of all marks on the qualification assess mathematical skills, and those marks are overwhelmingly clustered in the methods content. The good news is that research-methods and maths marks are the most predictable on the whole qualification: the questions recur in fixed forms ("identify the design", "name and justify a statistical test", "calculate the percentage", "interpret this result"), and the technique for each is learnable. This lesson teaches that technique with worked numerical examples and KaTeX formulae, plus a decision procedure for choosing between the five Edexcel inferential tests and banded model answers showing what full-mark working looks like. One thing to fix at the outset: Edexcel examines a specific set of five non-parametric tests — there are no t-tests and no Pearson's r on this specification, a genuine difference from some other boards.

---

Spec Mapping

Edexcel 9PS0 — Exam Preparation. This is a synthesis lesson mapping the methods and mathematical skills that underpin the assessment.

Skill cluster	AO	Where assessed
Experimental methods, designs, sampling, variables	AO1 / AO2 / AO3	Paper 3; embedded in study questions on Papers 1 & 2
Descriptive statistics and data handling	AO2 (maths)	Paper 3; data items anywhere
Choosing and justifying the five inferential tests	AO2	Paper 3
Significance, critical values, Type I/II errors	AO1 / AO2	Paper 3
Percentages, fractions, ratios, sig figs, standard form, SD interpretation	AO2 (maths)	Minimum 10% of marks across all papers
Reliability, validity, ethics, peer review	AO1 / AO3	All papers

Assessment Objectives. Methods items are typically AO2-dominated (apply a concept to a described study; carry out a calculation) and AO3-weighted (evaluate a design, sample or ethical safeguard in context), on an AO1 base of accurate definitions. Because most methods questions are set around a novel-scenario stem, scenario-free answers score poorly.

Papers this supports. Principally Paper 3, but methods reasoning is examinable on any study across all three papers.

Key Point: Treat research methods as a cross-cutting skill, not a Paper 3 silo. A "design an element of this study" or "identify a confounding variable" question can attach to any described study on any paper, and the mathematical minimum is met partly through these embedded items.

---

Experimental Methods

Type	Description	Strengths	Limitations
Laboratory experiment	Controlled environment; researcher manipulates the IV, measures the DV	High control; easy to replicate; can establish cause and effect	Low ecological validity; demand characteristics; artificial
Field experiment	Natural setting; researcher still manipulates the IV	Higher ecological validity; more natural behaviour	Less control; harder to replicate; ethical issues (no consent)
Natural experiment	The IV occurs naturally (not manipulated); the DV is measured	Allows study of variables unethical/impractical to manipulate	No cause and effect; confounding variables
Quasi-experiment	IV is an existing participant characteristic (age, gender, diagnosis)	Allows comparison of pre-existing groups	No random allocation; differences may be confounds

---

Experimental Designs

The design determines how participants are allocated to conditions.

Design	Each participant...	Strengths	Limitations	Fix
Independent groups	takes part in ONE condition	No order effects; less likely to guess the aim	Individual differences confound; needs more participants	Random allocation
Repeated measures	takes part in ALL conditions	Controls individual differences; fewer participants	Order effects; demand characteristics	Counterbalancing (ABBA)
Matched pairs	is paired then split across conditions	Reduces individual differences; no order effects	Time-consuming; cannot match on everything	Match on the most relevant variable

Worked technique — "Identify the design": read the procedure and ask one question — did the same people do every condition? If yes, it is repeated measures (or, if pre-paired, matched pairs); if different people did each condition, it is independent groups. State the design and one justification drawn from the stem.

---

Sampling Methods

The sample is drawn from the target population; the method determines representativeness.

Method	How it works	Strengths	Limitations
Random	Every member has an equal chance (random-number generator)	Free from researcher bias; likely representative	Time-consuming; selected people may decline
Systematic	Every nth person on a list	Objective; easy	A periodic pattern in the list can bias it
Stratified	Population split into strata; randomly sampled in proportion	Highly representative of key characteristics	Very time-consuming; needs detailed population data
Opportunity	Whoever is available and willing	Quick, easy, cheap	Highly biased; over-represents one group
Volunteer	Participants self-select (advert)	Willing participants; good for sensitive topics	Volunteer bias (more motivated/extravert)

Exam Tip: When evaluating a sampling method, always address (a) representativeness of the target population and (b) any systematic bias the method introduces — and do so in the context of the described study, naming the population and the likely biased subgroup.

---

Non-Experimental Methods

Not every study is an experiment. Edexcel examines several non-experimental techniques, and a common question asks you to evaluate or design an element of one.

• Observations — naturalistic vs controlled, covert vs overt, participant vs non-participant. Made systematic using behavioural categories (operationalised, observable actions) and sampling such as event sampling (count each occurrence) or time sampling (record at fixed intervals). Inter-observer reliability is checked by correlating two observers' records.
• Self-report — questionnaires and interviews (structured, semi-structured, unstructured). Strengths include large samples (questionnaires) and rich detail (unstructured interviews); limitations include social desirability bias and response sets. Good design avoids leading and double-barrelled questions.
• Correlations — examine the relationship between two co-variables without manipulating an IV, so they cannot establish cause and effect. The classic evaluation point is the third-variable problem.
• Case studies and content analysis — in-depth study of an individual or group, and the systematic analysis of existing material into categories or themes.

---

Reliability and Validity

These two concepts are examinable in their own right and provide ready-made AO3 across the whole course.

Concept	Question it answers	Types / checks
Reliability	Is the measure consistent?	Test-retest (same test, two occasions); inter-observer (two observers agree)
Validity	Does it measure what it claims?	Internal (free of confounds); external (ecological, population, temporal); face and concurrent validity

Ways to improve reliability include standardising procedures, operationalising behavioural categories and training observers. Ways to improve validity include controlling extraneous variables, using a control group, and checking a new measure against an established one (concurrent validity). Remember: reliability is necessary but not sufficient for validity — a measure can be perfectly consistent and still measure the wrong thing.

---

Descriptive Statistics (with Worked Maths)

Measures of Central Tendency

Measure	Method	Strength	Limitation
Mean	sum ÷ number of values	Uses all data; most sensitive	Distorted by outliers
Median	middle value when ordered	Unaffected by outliers	Ignores most data
Mode	most frequent value	Works for categorical data	May not exist / may be multiple

The mean of a data set is defined as:

$\bar{x} = \frac{\sum x}{n}$xˉ=n∑x​

where $\sum x$∑x is the sum of all scores and $n$n is the number of scores. For the data set 4, 7, 7, 9, 13:

$\bar{x} = \frac{4 + 7 + 7 + 9 + 13}{5} = \frac{40}{5} = 8$xˉ=54+7+7+9+13​=540​=8

The median is the middle value (7) and the mode is the most frequent value (7).

Measures of Dispersion

Measure	Method	Strength	Limitation
Range	highest − lowest (sometimes +1)	Quick	Uses only two values; sensitive to outliers
Standard deviation	average distance of scores from the mean	Uses all data; precise	Harder to compute; affected by outliers

The standard deviation measures the average spread of scores around the mean. One common form is:

$\sigma = \sqrt{\frac{\sum (x - \bar{x})^2}{n}}$σ=n∑(x−xˉ)2​​

Interpreting it is the examinable skill: a large standard deviation means scores are widely spread from the mean (more variability); a small standard deviation means scores cluster tightly around the mean (more consistency). For the data above ($\bar{x} = 8$xˉ=8), the squared deviations are $16, 1, 1, 1, 25$16,1,1,1,25, giving:

$\sigma = \sqrt{\frac{44}{5}} = \sqrt{8.8} \approx 2.97$σ=544​​=8.8​≈2.97

Exam Tip: You must be able to calculate the mean, median, mode and range. For standard deviation, Edexcel expects you to understand and interpret it (high SD = spread out; low SD = clustered) and to recognise the formula, rather than to compute it unaided under exam conditions.

Levels of Measurement

Every test-choice decision starts here, so learn the three levels cold.

Level	Description	Example
Nominal	Named categories; frequency counts	How many chose A vs B
Ordinal	Can be ranked; unequal intervals	Race positions; Likert ratings
Interval	Equal intervals on a scale	Time in seconds; standardised test scores

---

The Five Inferential Tests: Choosing the Right One

Edexcel 9PS0 examines a fixed set of five non-parametric tests: the sign test, Mann-Whitney U, Wilcoxon signed-rank, Spearman's rho and chi-square. Choosing between them is not a matter of memory or intuition — it is a mechanical decision fixed by three questions, asked in order:

1. Hypothesis — a difference or a correlation/association?
2. Design — related (repeated measures / matched pairs) or unrelated (independent groups)?
3. Level of measurement — nominal, ordinal or interval?

Because all five tests are non-parametric, interval data are analysed using the ordinal tests (Wilcoxon, Mann-Whitney, Spearman) unless a question states otherwise — the rank-based tests are perfectly valid on interval data, they simply convert scores to ranks first.

Decision Tree

graph TD
    A["What is the hypothesis?"] --> B["Difference"]
    A --> C["Correlation / association"]
    B --> D{"Related or unrelated?"}
    D --> E["Related design"]
    D --> F["Unrelated design"]
    E --> G{"Level of data?"}
    F --> H{"Level of data?"}
    G --> I["Nominal: Sign test"]
    G --> J["Ordinal / interval: Wilcoxon"]
    H --> K["Nominal: Chi-square"]
    H --> L["Ordinal / interval: Mann-Whitney U"]
    C --> M{"Level of data?"}
    M --> N["Ordinal / interval: Spearman's rho"]
    M --> O["Nominal: Chi-square (association)"]

Summary Table

Test	Difference or Correlation	Design	Level of Data
Sign test	Difference	Related	Nominal (direction of change)
Wilcoxon signed-rank	Difference	Related	Ordinal / interval
Mann-Whitney U	Difference	Unrelated	Ordinal / interval
Chi-square ($\chi^2$χ2)	Difference / association	Unrelated	Nominal (frequencies)
Spearman's rho ($r_s$rs​)	Correlation	N/A	Ordinal / interval

Exam Tip: The stock question — "Name a suitable statistical test and justify your choice" — wants the test plus all three justifications: the hypothesis (difference/correlation), the design (related/unrelated), and the level of data. Naming the test alone rarely earns full marks; the justification is where the credit lies. A strong answer also eliminates the nearest alternatives ("not Mann-Whitney, because the design is related, not independent").

---

Probability and Significance

The Significance Level

The conventional significance level in psychology is $p \leq 0.05$p≤0.05 (5%). This means:

• There is a 5% or smaller probability that the result occurred by chance, if the null hypothesis were true.
• If $p \leq 0.05$p≤0.05, the result is statistically significant and the null hypothesis is rejected.
• If $p > 0.05$p>0.05, the result is not significant and the null hypothesis is retained.

A more stringent level such as $p \leq 0.01$p≤0.01 is used where a Type I error would be especially costly (for example, in drug or medical research).

Type I and Type II Errors

Error	What happens	More likely when	Nickname
Type I	Reject a true null hypothesis (false positive)	Significance level too lenient (e.g. $p \leq 0.10$p≤0.10)	Optimistic error
Type II	Retain a false null hypothesis (false negative)	Significance level too stringent (e.g. $p \leq 0.01$p≤0.01)	Pessimistic error

Key Point: $p \leq 0.05$p≤0.05 is a deliberate compromise. Relaxing to $p \leq 0.10$p≤0.10 raises the Type I risk; tightening to $p \leq 0.01$p≤0.01 raises the Type II risk. There is no level that minimises both at once.

The two errors become intuitive with a concrete example. Imagine a study testing whether a new revision technique improves recall, where in reality it makes no difference at all. A Type I error would occur if the sample happened, by chance, to show a large improvement, the analysis returned a significant result, and the researcher wrongly concluded the technique works — claiming an effect that is not really there. A Type II error would occur in the mirror-image situation: the technique genuinely does improve recall, but the particular sample, perhaps because it was small or unusually variable, failed to reach significance, so the researcher wrongly concluded there was no effect — missing a real effect. The reason the two cannot be minimised together is that they pull in opposite directions: making the test harder to pass (a stricter level such as $p \leq 0.01$p≤0.01) means you are less likely to be fooled by a chance result (fewer false positives), but you are also more likely to overlook a genuine but modest effect (more false negatives). Because a Type I error means publishing a false claim, which can mislead an entire field, psychology treats it as the more serious of the two and fixes the conventional level at 5% rather than something more relaxed. In an exam, the discriminator between a mid-band and a top-band answer is almost always whether the candidate ties each error to what the researcher actually concluded about the null hypothesis, rather than reciting "false positive / false negative" as bare labels.

Interpreting a Significant Result in Words

Once a test has been run, the examinable skill is often to interpret the outcome in plain English rather than to produce the number. Suppose a Mann-Whitney U test comparing recall under quiet versus noisy conditions gives a calculated value that is smaller than the critical value at $p \leq 0.05$p≤0.05 for a one-tailed test. A full interpretation has three moving parts, and a strong answer states all three. First, the statistical decision: because the calculated value is less than or equal to the critical value, the result is significant, so the null hypothesis is rejected and the alternative (experimental) hypothesis is accepted. Second, the meaning in context: participants in the quiet condition recalled significantly more words than those in the noisy condition, and this difference is unlikely to be due to chance. Third, the confidence and its limits: "significant at $p \leq 0.05$p≤0.05" means there is a 5% or smaller probability the result arose by chance if the null were true — it does not mean the effect is large, important, or certain, and there remains that small residual risk of a Type I error. A common and costly slip is to leap from "significant" to "this proves the technique causes better memory" — significance tells you an effect is unlikely to be chance, not that it is large or that causation is established beyond the design's own limits. Where the question also supplies descriptive statistics, the best answers weave them in: the direction of the means confirms which condition performed better, while the inferential test confirms the difference is reliable, so the two kinds of statistic do different jobs and should be reported together rather than in place of one another.

Comparing the Calculated and Critical Value