Normal Distribution
A continuous probability distribution that is symmetric, bell-shaped, and fully characterised by two parameters: the mean μ (location) and the standard deviation σ (spread). The probability density is highest at the mean and decreases smoothly as values move further away in either direction. Also called the Gaussian distribution after C. F. Gauss, and the basis for most parametric inferential statistics.
Properties
A normal distribution is symmetric about its mean, with mean, median, and mode coinciding. Roughly 68% of the probability mass lies within μ ± σ, 95% within μ ± 1.96σ, and 99.7% within μ ± 3σ. Standardising by computing z = (x − μ) / σ converts any normal variable to the standard normal distribution with μ = 0 and σ = 1, allowing tabulated probabilities to be applied to any normally distributed measure.
Central Limit Theorem
The central role of the normal distribution in inferential statistics rests on the central limit theorem: the sampling distribution of the mean of independent observations approaches normality as sample size grows, regardless of the shape of the underlying population, provided the population variance is finite. This justifies normal-approximation confidence intervals and z-tests for means even when raw scores are non-normal, once samples are sufficiently large — a rough rule of thumb in applied work being N ≥ 30 per group.
Parametric Assumptions
t-tests, ANOVA, linear regression, and most other parametric techniques assume normally distributed residuals (not raw scores). Larson-Hall (2016) and Field (2018) both stress that mild departures from normality rarely affect inference when N is reasonably large, but with small classroom samples the assumption matters and should be checked using histograms, Q-Q plots, and tests of normality (Shapiro–Wilk, Kolmogorov–Smirnov), with non-parametric alternatives (Mann–Whitney U, Kruskal–Wallis, Wilcoxon) used when the assumption fails.
In Language Testing
Total scores on well-constructed standardised language tests are approximately normally distributed across large calibration samples, by design — items are calibrated and selected so that the resulting score distribution is roughly bell-shaped. Sub-skill scores at the extremes of the proficiency range often show floor or ceiling effects that violate normality.
References
- Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics (5th ed.). London: Sage.
- Larson-Hall, J. (2016). A Guide to Doing Statistics in Second Language Research Using SPSS and R (2nd ed.). New York: Routledge.
- Bachman, L. F. (2004). Statistical Analyses for Language Assessment. Cambridge: Cambridge University Press.