Data handling and analysis — AQA A-Level Psychology
Test yourself on Data handling and analysis with AQA A-Level practice questions.
7 days Premium · Then the free plan continues · No card, no charge
Data handling and analysis explained
A number is quantitative, a word is qualitative, and the technique you choose decides which you get.
Read the full explanation
Closed questions, rating scales, behavioural categories in a structured observation and reaction times all produce scores; open questions, unstructured interviews and case studies produce language. Milgram gives you both: 65% of participants going to 450 volts is quantitative, while his notes on grown men sweating, stuttering and laughing nervously are qualitative. Quantitative data is objective, quick to analyse and easy to compare across conditions, but it strips out meaning; qualitative data is rich and has greater external validity, yet what it means depends on the researcher reading it, so it invites bias and resists summary. Content analysis turns the second into the first.
Primary and secondary data, including meta-analysis.
Primary data is gathered first-hand by the researcher, for this investigation's own aim, through an experiment, questionnaire, interview or observation. Secondary data already exists: somebody else collected it, usually for another purpose, as with government statistics, medical records or a published study. Primary data fits the hypothesis exactly but costs time and money; secondary data is cheap and immediate, though it may be outdated, incomplete or produced under conditions you cannot check. A meta-analysis is the disciplined use of secondary data: results from many studies of one question are pooled and an effect size, a standardised measure of how strong the finding is, lets studies of different sizes be compared. Van IJzendoorn and Kroonenberg pooled 32 Strange Situation studies and found variation within countries larger than between them.
Descriptive statistics: measures of central tendency – mean, median, mode; calculation of mean, median and mode; measures of dispersion; range and standard deviation; calculation of range; calculation of percentages; positive, negative and zero correlations.
Central tendency asks where scores cluster, dispersion asks how far they spread. The mean uses every score, making it sensitive but easily distorted by anomalies. The median, the middle score of ordered data, shrugs off anomalies. The mode is the most frequent score. Range is the highest score minus the lowest. Standard deviation is the average distance of scores from the mean; a large one means the group is varied. A percentage is the part over the whole, multiplied by 100. Correlations show relationships between co-variables: positive means as one increases, the other increases; negative means as one increases, the other decreases; zero means no relationship exists.
Presentation and display of quantitative data: graphs, tables, scattergrams, bar charts, histograms.
A table in a report is not a list of raw scores; it is a summary, giving the descriptive statistics for each condition under a title that names the variables and the units. Bar charts display data in categories or conditions, and their bars carry gaps because there is no continuous scale along the bottom, while the vertical axis shows frequency or a mean. Histograms display continuous data, so the bars touch, the horizontal axis is divided into equal intervals and no interval is skipped. A scattergram displays a correlation: one co-variable on each axis, one dot per participant, and the shape of the cloud tells you direction and strength. Label both axes and give a title every time. Watch for a vertical axis that does not begin at zero, which makes a trivial difference look dramatic.
Distributions: normal and skewed distributions; characteristics of normal and skewed distributions.
A normal distribution is the symmetrical bell curve occurring when most people score near the middle and fewer score towards either extreme; height, shoe size and IQ approximate it. Its characteristics are: symmetrical about the midpoint, the mean, median and mode all falling at the same central point, and tails that stretch towards the horizontal axis without ever touching it, as extreme scores remain possible. Roughly 68% of scores sit within one standard deviation of the mean. A skewed distribution is asymmetrical and named after its tail. Positive skew has a long tail to the right, scores bunched at the low end and the mean pulled above the median and mode, as seen in a very difficult test; negative skew is its mirror image.
Analysis and interpretation of correlation, including correlation coefficients.
A correlation measures how two co-variables move together. Neither is manipulated, both are merely measured, which is exactly why a correlation can never on its own establish a cause. The coefficient is one number running from -1 through 0 to +1: its sign gives direction and its size gives strength, so a coefficient of -0.76 describes a relationship precisely as strong as +0.76, simply running the other way. A figure near zero means the two vary independently. Whether a coefficient is significant is decided against a critical value table and depends on sample size, so with a large enough sample even a feeble coefficient reaches significance and means little in practice. Two traps follow: a third variable may be driving both, and a curvilinear relationship yields a coefficient close to zero.
Levels of measurement: nominal, ordinal and interval.
Levels of measurement rank data by how much information each score carries, and the level decides what you may legitimately do with it. Nominal data is counted in categories: how many chose each brand, how many passed and how many failed. Every item falls in one category only, the categories have no order, and the mode is the only average available. Ordinal data can be placed in rank order, but the gaps between ranks are unequal and the units are not standardised; a mood rating out of ten is ordinal, because your seven and my seven are not the same thing, so the median is the safe average. Interval data uses equal, fixed units on a publicly agreed scale, such as seconds or words recalled, and only interval data earns the mean and standard deviation.
Coding in content analysis.
Coding is the initial stage of content analysis where large amounts of qualitative data, such as interview transcripts or media articles, are categorised into meaningful units. Because qualitative data is often extensive, researchers must establish a coding system to systematically identify specific concepts. For example, when analysing diaries for signs of depression, a researcher might create codes for words like 'hopeless' or 'exhausted'. By tallying the frequency of these codes, the researcher transforms rich qualitative information into quantitative data. This allows for easier comparison, identification of patterns, and subsequent statistical analysis, making the data more manageable and objective.
Your focus
- Distinguish quantitative from qualitative data and name the technique that produces each of them.
- Apply the distinction to a described study by saying which type it produced and quoting the detail that shows it.
- Evaluate the trade between objectivity and easy comparison on one side and depth of meaning on the other.
Show all 26 objectives
- Describe what makes data primary or secondary and define a meta-analysis and the effect size it produces.
- Apply the distinction to a described investigation by stating which type of data it used and how you can tell.
- Evaluate meta-analysis by setting its huge combined sample against publication bias inflating the pooled effect.
- Calculate the mean, median, mode and range for a set of scores and state what each figure says about them.
- Calculate a percentage as the part divided by the total and multiplied by 100, showing the working.
- Explain what a large standard deviation says about the spread and about how well the mean describes anyone.
- Explain why an anomalous score makes the median a safer average than the mean for a given set of results.
- Identify and explain positive, negative, and zero correlations from a description of co-variables or a scattergram.
- Describe when a bar chart, a histogram and a scattergram is each the correct display for the data.
- Sketch a labelled graph for given data with both axes named, units shown and a title naming the variables.
- Identify how a vertical axis that does not start at zero can make a trivial difference look dramatic.
- Describe a normal distribution by its symmetry, the mean median and mode at one point, and tails that never touch.
- Sketch a positively and a negatively skewed curve and mark where the mean, median and mode fall on each.
- Apply the idea to a set of results by saying which distribution a far too difficult test would produce and why.
- Explain how the sign and the size of a coefficient give the direction and the strength of a relationship.
- Interpret a stated coefficient in the terms of the two named variables and say whether it reached significance.
- Evaluate a correlational claim using the third-variable problem and the curvilinear case of arousal and performance.
- Describe nominal, ordinal and interval data by how much information each level of measurement carries.
- Identify the level of a described measure and state which average is legitimate to use with it.
- Justify why interval data makes a test more powerful and less likely to miss a real effect.
- Define the process of coding within the context of content analysis.
- Explain how coding transforms qualitative data into quantitative data.
- Design a simple, operationalised coding system for a given piece of qualitative data.
Data handling and analysis exam tips
Marking Points
- Quantitative data is recorded as numbers; qualitative data is recorded as words, images or meanings.
- Naming the technique that produces each: closed questions, structured observation and experiments give quantitative data, while open questions, unstructured interviews and case studies give qualitative data.
- Strength of quantitative data: it is objective, can be summarised with descriptive statistics and analysed with an inferential test, so conditions can be compared directly.
- Strength of qualitative data: it captures detail and meaning the researcher never anticipated, which raises external validity.
- The matching limitations stated as consequences: a score can misrepresent the experience behind it, and qualitative material has to be interpreted, so the conclusion may carry the researcher's bias.
- Applying the distinction to the stem by saying which type the study produced and quoting the detail that shows it.
- Primary data: collected first-hand by the researcher for the current aim.
- Secondary data: already in existence, collected by another person, often for a different purpose.
- Strength of primary data: it targets the hypothesis, and the researcher controls the conditions under which it was produced.
- Strengths of secondary data: inexpensive, immediately available and often drawn from far larger samples than one researcher could gather.
- Meta-analysis defined as the statistical pooling of results from many studies of the same question, with an effect size calculated as a standardised measure of the strength of the finding.
- Evaluation of meta-analysis: the combined sample is huge, so generalisability improves, but publication bias means non-significant studies are less likely to be in print and the pooled effect can be inflated.
- Mean: add every score and divide by how many there are, so it uses all the data and is the most representative measure when there are no anomalies.
- Median: put the scores in order and take the middle one, or the midpoint of the middle two when the number of scores is even.
- Mode: the most frequently occurring score, and the only measure of central tendency available for data in named categories.
- Range: subtract the lowest score from the highest, a quick measure of dispersion that uses only two of the values.
- Standard deviation: the average amount by which scores deviate from the mean, a larger figure meaning greater spread and a less representative mean.
- Percentage: the part divided by the total and multiplied by 100, with the working shown.
- Correlation direction: positive (both variables increase together), negative (one increases as the other decreases), or zero (no relationship).
- Choosing the right display for the data: bar chart for categories or conditions, histogram for continuous data grouped into intervals, scattergram for a correlation.
- Gaps between the bars of a bar chart because the categories are discrete, and touching bars on a histogram because the underlying scale is continuous.
- Both axes labelled, units included, and a title that names the variables.
- A summary table presents descriptive statistics rather than raw data, and is read with a sentence saying what it shows.
- Reading a scattergram: an upward pattern is a positive correlation, a downward pattern negative, and a formless scatter of points a zero correlation.
- Evaluation: a graph can mislead, for instance where the vertical axis does not start at zero, so an apparently large difference may be trivial.
- Normal distribution described as symmetrical and bell-shaped, with most scores concentrated around the mean.
- In a normal distribution the mean, median and mode all fall at the same central point.
- The tails approach but never meet the horizontal axis, so extreme scores are rare rather than impossible.
- Skewed distribution defined as asymmetrical, with the name taken from the direction of the long tail.
- Positive skew: long tail to the right, scores clustered at the lower end, mode lowest and mean highest.
- Negative skew: long tail to the left, scores clustered at the upper end, mean lowest and mode highest.
- An example that explains the shape, such as a test that was too hard for the sample producing positive skew.
- Correlation defined as an association between two co-variables, both of which are measured rather than manipulated.
- Direction read from the sign: positive means both rise together, negative means one rises as the other falls, zero means no relationship.
- Strength read from how close the figure lies to one, independently of its sign.
- Interpretation set in the stem, stating what the relationship means for the two named variables.
- Significance judged against a critical value, with sample size affecting whether a given coefficient counts as significant.
- Evaluation: a correlation cannot demonstrate causation, because an untested third variable may be responsible for both co-variables.
- Evaluation: a curvilinear relationship, such as arousal and performance, returns a coefficient near zero despite a strong systematic relationship.
- Nominal: frequency counts in named categories that have no order, with each score placed in one category only.
- Ordinal: scores can be ranked, but the intervals between them are unequal and the scale rests on subjective judgement.
- Interval: equal intervals on a standardised, publicly agreed scale, such as time in seconds or temperature.
- The level determines the appropriate average: mode for nominal, median for ordinal, mean for interval.
- The level is one of the three things that choose an inferential test, alongside whether the study tests difference or correlation and whether the design is related or unrelated.
- Evaluation: interval data carries the most information, so tests using it are more powerful and less likely to miss a real effect.
- Define coding as the process of categorising qualitative data into manageable, meaningful units.
- Explain that coding typically transforms qualitative data into quantitative data through the use of frequencies or tallies.
- Describe the necessity of establishing clear, operationalised coding categories before or during the analysis of the data.
- Outline how researchers apply the coding system systematically across the entire dataset to ensure consistency and reliability.
Examiner Tips
- 💡When the question says 'from the stem', quote the detail that settles it: a rating scale out of ten is quantitative, an open 'describe how you felt' is qualitative.
- 💡For a two-mark 'explain one limitation', name the limitation in the first sentence and spend the second on what it does to the conclusion.
- 💡Where an item asks you to distinguish between the two, write about both inside one sentence; a paragraph on each often never states the difference at all.
- 💡Settle primary versus secondary on who collected it and for what aim, not on how old the data is.
- 💡For a four-mark 'outline and evaluate meta-analysis', one clean definition plus one developed strength and one developed limitation beats four undeveloped points.
- 💡Name the evaluative problem, such as publication bias or differences in method between the pooled studies, rather than writing that the findings 'might not be accurate'.
- 💡Calculation marks are awarded for method as well as answer, so set the working out; a wrong final figure with visible correct method can still score.
- 💡If the data contains an obvious outlier, justify the median and say why: it is unaffected by the extreme score.
- 💡Standard deviation items nearly always ask what the figure tells you about the participants, so finish with a sentence on how consistent or varied the group was.
- 💡When interpreting a scattergram, explicitly state whether the correlation is positive, negative, or zero based on the direction of the data points.
- 💡Read the level of measurement in the stem before choosing the graph; categories mean a bar chart whatever the study is about.
- 💡'Sketch' still requires labelled axes and a title, but the plotting need not be to scale.
- 💡When asked what a graph shows, describe the pattern using the stem's variables rather than saying one bar is taller than another.
- 💡Sketch it. A labelled curve with the mean, median and mode marked in the right order earns marks that prose struggles to.
- 💡If you cannot recall which skew is which, draw the tail first and read the name off the side it points towards.
- 💡Where a stem hands you a mean well above the median, that is the examiner signalling a positively skewed distribution.
- 💡Turn the coefficient into a sentence about the stem's own variables before you evaluate it.
- 💡If asked why a conclusion is unjustified, name a plausible third variable drawn from the scenario instead of stopping at 'correlation is not causation'.
- 💡Quote the figure when you evaluate; 'a strong correlation' without the number leaves the interpretation mark unclaimed.
- 💡Ask where the numbers came from: a stopwatch or a count of words is interval, a rating scale or a placing is ordinal, a tally of categories is nominal.
- 💡When asked to identify and justify a level of measurement, name the level and use because to link it to how the data were produced.
- 💡State the level before naming the test; examiners credit the reasoning chain, not just the test's name.
- 💡When asked to design a content analysis, explicitly state the specific codes you would use for the given scenario.
- 💡Mention how inter-rater reliability can be established by having two independent researchers apply the same coding system and correlating their tallies.
- 💡Clearly distinguish between the qualitative input (the raw data) and the quantitative output (the tallied codes) when explaining the process.
Common Mistakes
- Confusing quantitative with quantity, so a long interview transcript is called quantitative because there is a great deal of it.
- Claiming qualitative data cannot be analysed, when content analysis and thematic analysis exist precisely to analyse it.
- Offering 'it is more detailed' as a strength without saying what the detail buys, which only earns the mark once the consequence for validity is stated.
- Treating the two as rival loyalties rather than tools, when many studies deliberately collect both.
- Labelling every published study as secondary data; a study is primary data for the team that ran it and secondary only for the researcher who later reuses it.
- Defining meta-analysis as 'a review of lots of studies', which describes a literature review; the defining feature is that the results are combined and analysed statistically.
- Omitting effect size altogether, which is the term the examiner is looking for.
- Asserting that secondary data is always less valid, when official statistics are often collected far more rigorously than a single researcher could manage.
- Giving a mean to more decimal places than the raw scores justify, or ignoring the rounding instruction printed in the question.
- Choosing a median without first placing the scores in order, which returns the middle score of an unordered list.
- Reading a large standard deviation as meaning the scores are higher, when it means they are more spread out.
- Describing a zero correlation as 'no result' rather than as a genuine finding of no relationship between the co-variables.
- Drawing a bar chart with touching bars, or a histogram with gaps, which forfeits the mark that distinguishes the two.
- Plotting two means as a bar chart when the stem describes a correlation, which needs a scattergram.
- Leaving the axes unlabelled, which is the most commonly lost mark on any drawing item.
- Copying every participant's raw score into what the question called a summary table.
- Naming the skew after where the bulk of scores sit rather than after the tail, which reverses positive and negative every single time.
- Drawing a skewed curve but leaving the three averages together in the middle, when the whole point of the diagram is that they separate.
- Writing that the tails touch the horizontal axis, which contradicts the idea that extreme scores remain possible.
- Incorrectly citing reaction times as normally distributed, when they are typically positively skewed.
- Reading a negative coefficient as a weak one, so that -0.8 is described as weaker than +0.3.
- Concluding that one variable causes the other, which is the most heavily penalised error on this topic.
- Calling the two measures independent and dependent variables; in a correlation they are co-variables.
- Treating a coefficient of zero as a failed study rather than as a genuine finding of no relationship.
- Classifying anything numerical as interval, when a questionnaire total out of twenty is ordinal.
- Reading 'nominal' as meaning small numbers rather than named categories.
- Saying ordinal data cannot be analysed, when it is ranked for tests such as Wilcoxon and Spearman's.
- Calculating a mean from a set of subjective ratings, which assumes an equality of intervals the scale does not possess.
- Confusing coding with thematic analysis. Correction: Coding typically generates quantitative data by counting frequencies, whereas thematic analysis keeps data qualitative by identifying broader descriptive themes.
- Failing to operationalise categories. Correction: Ensure coding categories are explicitly defined so that independent researchers would code the data identically, ensuring inter-rater reliability.
- Assuming coding only applies to written text. Correction: Coding can be applied to various media, including audio recordings, video footage, and visual advertisements.