Research in Educational Psychology

This is Unit Two, on research in educational psychology: why teaching needs evidence, how researchers gather and measure it, and what each design lets a study claim. Unit One closed on matching a classroom to its philosophy. Unit Two asks how anyone knows which classroom practices work. Carry one idea through this unit: a study may claim only what its design licenses. At the entry on randomness, the unit shows the two uses of chance that decide who a finding reaches and whether it shows cause. By the end you will be able to explain why teaching needs evidence rather than impression. To tell reliability from validity. To read a correlation's sign, size and limit. To say what random sampling and random assignment each license. And to choose a developmental design for a question.

Four entries on why teaching needs evidence, and what makes an inquiry scientific.

Why teaching needs evidence.

EDP 0101 Why Teaching Needs Evidence

Teaching has often been treated as an art and shaped by ideology; evidence-based teaching asks what research shows actually works in classrooms. Goldacre urged giving teachers access to research and training them to read it critically. And a finding that changes with new evidence is science working, not failing.

Why personal experience misleads.

EDP 0102 Confirmation Bias

Learning from experience and from what others agree on is unreliable because of confirmation bias: seeking evidence we are right and ignoring the rest. A teacher sure that Monday lessons go badly notices every rough Monday and forgets the smooth ones. That is why science tests claims instead of trusting impressions.

The test every scientific theory must pass.

EDP 0103 Falsifiability

Popper argued that a theory is scientific only if it is falsifiable: some possible observation could show it wrong. A claim that every child learns best in some hidden way no test detects can never be refuted, so science cannot test it. Falsifiable does not mean false: a theory can survive every test so far and still count.

Three features of science.

EDP 0104 Three Features of Science

Stanovich names three features of science: systematic empiricism, planned and recorded observation; empirical questions, answerable by observing the world; and public knowledge, shared through publication so others can check, repeat and correct it. Research runs as a cycle, from question to publication and on to the new questions the answers raise.

Eight entries on the kinds of research, the ways to collect data, and how to measure what cannot be seen.

Research that counts, and research that describes.

EDP 0105 Quantitative and Qualitative Research

Quantitative research gathers a little data from many people and uses statistics to reach general conclusions, like a survey of two thousand students; qualitative research gathers rich data from a few, describing lived experience, like long interviews with ten. Quantitative work is precise but poor at raising new questions; qualitative work generates hypotheses.

Studies that use both.

EDP 0106 Mixed Methods and Triangulation

Mixed-methods research combines the two, using qualitative work to generate hypotheses and quantitative work to test them, as when interviews suggest a pattern and a survey checks it across the school. Running both at once is triangulation: agreement strengthens a finding, and disagreement opens a new question.

The three main kinds of research.

EDP 0107 Three Kinds of Research

Research falls into three main kinds: descriptive, which reports a variable, such as a count of absences; correlational, which tests whether variables are related, such as absences against grades; and experimental, which tests cause, such as an attendance program tried at random. Only the experiment, with random assignment, can show cause.

Watching behavior directly.

EDP 0108 Observation

Observation records behavior directly, either naturalistically, in its natural setting without control, or in a laboratory, where confounds can be controlled at a cost. An observer might tally how often students ask questions in ordinary lessons. But people change their behavior when watched, the Hawthorne effect, which laboratory observation cannot remove.

Asking people about themselves.

EDP 0109 Surveys and Self-Report

Surveys collect many variables quickly, through forced-choice items that are fast to code or open-ended items that go deeper but are hard to compare. Self-report is limited by image management, discomfort and poor self-knowledge, while archives and existing data sets avoid recruiting but hold only what was collected.

Measuring what cannot be seen.

EDP 0110 Tests and Constructs

Tests measure constructs, traits, states, attitudes and abilities that cannot be observed directly, as a self-esteem scale or an aptitude test does. A good test is standardized: administered, scored and analyzed identically for everyone, which is what lets one person's score be compared fairly with another's.

Three ways a test can be consistent.

EDP 0111 Three Kinds of Reliability

Reliability is consistency, checked three ways: test-retest over time, internal consistency across a test's items, and interrater reliability across the people scoring it, as when two teachers give the same essays nearly the same scores. Each answers a different question, so a test can be stable yet have items that disagree.

Why consistency is not enough.

EDP 0112 Reliable but Not Valid

Validity is whether scores represent the variable a test intends to measure. A valid measure must be reliable, but a reliable one need not be valid: a clock set ten minutes fast is perfectly consistent and consistently wrong. A math test full of hard reading may measure reading as much as math.

Seven entries on statistics: distributions, significance, samples, and reading correlations without mistaking them for causes.

Looking past the average.

EDP 0113 Distributions, Not Only Averages

Statistics turns data into insight, and its core habit is distributional thinking: data vary, and the whole distribution often reveals more than the center. Two classes can share a median of seventy, one bunched and one spread wide. Number-crunching is a small part; planning and honest conclusions matter as much.

What statistical significance means.

EDP 0114 What a P-Value Says

A p-value is the probability of results at least this extreme if chance alone were at work; by convention, under point zero five counts as significant. A gain that chance alone would produce fewer than once in a hundred tries is significant. It is not the probability that the hypothesis is true.

How far a sample's result can be trusted.

EDP 0115 Samples and Margin of Error

Results generalize from a sample to a population only when the sample is random, and the margin of error shows how far a sample's result might miss; polling four times as many randomly chosen students roughly halves it. The margin covers chance in sampling, not bias from convenience samples or dishonest answers.

The number that measures a relationship.

EDP 0116 The Correlation Coefficient

A correlation coefficient, r, runs from minus one to plus one: its size gives the strength of a relationship and its sign the direction. Height and weight rise together, tiredness falls as sleep rises, and shoe size and sleep are unrelated. Correlations allow prediction, as admissions offices use scores and grades.

Reading a correlation step by step.

EDP 0117 Reading a Correlation

To read a correlation, take three steps. First the sign: plus means the variables rise together, minus means one rises as the other falls. Second the size: the closer r is to one, in either direction, the stronger the relationship, and values near zero mean little or none. Third the limit: a correlation never shows cause. Days of short sleep and grade point average correlate at minus point two nine: more short nights go with lower grades, weakly. And minus point eight is a strong relationship running the other way, not the absence of one.

The most common error in reading research.

EDP 0118 Correlation Is Not Causation

Correlation does not show causation, because a confounding variable may drive both: ice cream sales and crime rise together, and both follow hot weather. If students who eat breakfast earn higher grades, family routines might explain both. Only an experiment, with random assignment, can establish that one variable causes another.

Seeing relationships that are not there.

EDP 0119 Illusory Correlation

An illusory correlation is a belief that two things are related when they are not, such as the idea that a full moon makes students wilder; a review of about forty studies found no link. Confirmation bias and vivid cases keep such beliefs alive, and illusory correlations help sustain prejudice.

Six entries on experiments, ending with the payoff: what a study's use of chance lets it claim.

The design that can test cause.

EDP 0120 The Experiment

An experiment tests a specific hypothesis: an experimental group receives the manipulation, a control group does not, and both are otherwise treated the same, as when one class practices with flashcards and another reviews as usual before the same quiz. Each variable needs an operational definition stating exactly how it is measured.

What is changed, and what is measured.

EDP 0121 Independent and Dependent Variables

The independent variable is the one the researcher manipulates; the dependent variable is the one measured to see whether it changed. Flashcards or not is independent; the quiz score is dependent. The dependent variable is the outcome that may depend on the independent one, which keeps the two apart.

Keeping expectations out of the results.

EDP 0122 Blinding and the Placebo Effect

In a single-blind study participants do not know their condition; in a double-blind study neither participants nor experimenters know. Students told a sugar pill aids focus may report better focus, the placebo effect. Blinding guards against that and against experimenter bias, where researchers' expectations shape what they record.

Two uses of chance, often confused.

EDP 0123 Random Sampling vs Random Assignment

Random sampling gives everyone in a population an equal chance of being chosen; random assignment gives every participant an equal chance of either condition. Drawing names district-wide is sampling; a coin flip for each student's condition is assignment. Sampling decides who results apply to; assignment decides whether cause can be claimed.

Comparisons without random assignment.

EDP 0124 Quasi-Experiments

A quasi-experiment compares groups the researcher cannot assign at random, such as boys and girls, or two teachers' existing classes whose students chose their sections. It looks like an experiment, with independent and dependent variables, but without random assignment a confound can explain the result, so it cannot support a claim about cause.

The payoff: what randomness licenses a study to claim.

EDP 0125 What Randomness Licenses

Here is the unit's governing idea: a study may claim only what its design licenses. Random sampling decides who the findings apply to, generalized to a population within a margin of error. Random assignment decides whether cause can be claimed, because chance balances known and unknown differences across the groups. Correlational studies cannot claim cause, and neither can quasi-experiments, which compare groups that existed beforehand. Take a district reporting that students in a new reading program scored higher. Were the schools a random sample? If not, the result may not reach beyond them. Were students assigned at random? If families chose the program, motivated families may explain the gain. And a p-value speaks only to chance: it is not the probability that the hypothesis is true. Significance speaks to chance, sampling to reach, assignment to cause.

Three entries on designs for studying development over time, and for studying a teacher's own practice.

Designs that follow development over time.

EDP 0126 Choosing a Developmental Design

Developmental research uses three designs. Cross-sectional research tests different age groups at one time: fast and cheap, but it shows age differences rather than change, and cohort effects can confound age. Longitudinal research measures the same people repeatedly: it shows change within individuals but is slow and costly, and the least healthy and least advantaged drop out most often. Sequential research follows several cohorts over time, separating age from cohort. To learn how reading speed changes from eight to twelve within the same children, choose longitudinal; for a quick snapshot of age differences, cross-sectional.

Research bounded in one place and time.

EDP 0127 Case Studies

A case study examines one bounded case, in a set place and time, in its natural context, drawing on multiple sources, like one school's first year of a new schedule traced through interviews, records and observation. It cannot be generalized, so good ones report their context in detail for readers to judge.

Teachers studying their own practice.

EDP 0128 Action Research

Action research is practitioners systematically investigating their own practice to improve it, in a cycle: identify a problem, define the purpose and consult the literature, plan an intervention, collect and analyze data, and plan the next intervention from what was learned. Reflection becomes research when it captures data and leads to a public output. A teacher whose homework rarely comes back reads studies on homework design, shortens assignments for a month and compares return rates. Like a case study, it improves one setting, and a before-and-after comparison without a control group leaves other causes open.

Look at what you can now do. You can explain why teaching needs evidence: experience is bent by confirmation bias, and science tests claims that could be shown wrong. You can tell reliability, consistency, from validity, measuring what is intended: the fast clock has one without the other. You can read a correlation by its sign, its size and its limit, since correlation never shows cause. You can say what chance licenses: random sampling decides who results reach, and random assignment whether cause can be claimed. And you can choose a developmental design: cross-sectional for a quick snapshot, longitudinal for change within the same people. Unit Three turns to the developing learner.