Research concepts

18 explained· 21 coming soon · See alsoAI Psychology Product

18Explained · every one with checked sources

Observer-Expectancy Effect
A research bias in which the expectations of the person running a study unintentionally change what participants do or how results are read.
Hawthorne Effect
The tendency for people to change their behaviour because they know they are being observed or studied, which can distort research results.
A/B Testing
A/B testing is a controlled experiment that randomly splits traffic between two versions of a product to measure which one changes a target metric.
ROUGE
A set of automatic scores that rate a generated summary by counting the words and word sequences it shares with summaries written by people, mainly measuring how much of the reference it covers.
BLEU Score
An automatic machine translation score that compares output with human reference translations by counting matching word sequences of up to four words, with a penalty for output that is too short.
Perplexity
The metric that measures how well a language model predicts a text: the exponential of its average negative log-probability per token.
AUC
The area under a classifier's ROC curve: the probability that the model scores a random positive example higher than a random negative one, across every possible cutoff.
ROC Curve
A graph of a classifier's true positive rate against its false positive rate at every decision threshold, showing how well it separates two classes.
Confusion Matrix
A table that counts a classifier's true positives, false positives, false negatives and true negatives, from which metrics like precision and recall are calculated.
F1 Score
The harmonic mean of precision and recall, a single score from 0 to 1 that is high only when a classifier finds most positives and few of its flags are wrong.
Recall
The share of actual positive cases that a classifier or search system finds: true positives divided by all real positives, so a miss lowers it and a false alarm does not.
Occam's Razor
Occam's razor is the rule that, between explanations fitting the same evidence equally well, the one assuming the fewest extra causes should be preferred until evidence says otherwise.
Correlation and Causation
The observation that two variables move together, and the separate, often false, claim that one of them causes the other.
Margin of Error
The range around a survey result, in percentage points, that is expected to contain the true population value at a stated confidence level.
Sampling
Sampling is choosing a subset of a population to study so the results can describe the whole group, using either random selection or convenience and judgment.
Thought Experiment
An imagined scenario reasoned through step by step to test an idea when a real experiment is impossible or unnecessary.
Voluntary bias
Voluntary bias, usually called volunteer bias, is a research error caused by people who choose to join a study differing from those who do not.
Survivorship Bias
A sampling error that comes from studying only the cases that survived a selection process and ignoring those that dropped out, so results look better than the full group.