Prediction Interval

{
  "id": "MLC-T-ENV-021908",
  "term": "Prediction Interval",
  "field": "Environment",
  "definition": "1. A type of statistical interval that has a specified probability (commonly 95%) of enclosing the value of a \"future\" unit that is predicted based on the available data and does not belong to the sample used to generate the prediction, assuming that the future unit is drawn from the same population. In the context of regression analysis, standard prediction intervals relate to uncertainty when the fitted regression is used to predict the response (Y) variable for specific values of X variables. In cases of linear regression with a single X variable, prediction intervals associated with all X values are conventionally depicted as a band enclosing the fitted regression line, bounded by two curves that diverge as X increases in distance from the mean X, in either direction. Prediction intervals address both unit variation (e.g., as evaluated using sample quantiles) and statistical error in estimating unknown population parameters (e.g., in estimating a regression slope and intercept) and therefore can be distinguished from confidence intervals which only address the statistical error in parameter estimation.\n\n2. Type of statistical interval that has a specified probability (commonly 95%) of enclosing the value of a \"future\" unit that is predicted based on the available data and does not belong to the sample used to generate the prediction, assuming that the future unit is drawn from the same population. In the context of regression analysis, standard prediction intervals relate to uncertainty when the fitted regression is used to predict the response (Y) variable for specific values of X variables. In cases of linear regression with a single X variable, prediction intervals associated with all X values are conventionally depicted as a band enclosing the fitted regression line, bounded by two curves that diverge as X increases in distance from the mean X, in either direction. Prediction intervals address both unit variation (e.g., as evaluated using sample quantiles) and statistical error in estimating unknown population parameters (e.g., in estimating a regression slope and intercept) and therefore can be distinguished from confidence intervals which only address the statistical error in parameter estimation.",
  "subject": "Information management and systems",
  "references": [
    "CADDIS Glossary",
    "CADDIS"
  ],
  "url": "https://mlchart.com/terminology/environment/prediction-interval/"
}