Glossary

Corrects R² to account for the degrees of freedom in the data.

Often called the Box-Jenkins forecasting methodology, ARIMA is a set of univariate time-series forecasting methods. ARIMA involves the identification, selection, and checking of models with estimated autoregressive (AR), integration or differencing (I), and moving average (MA) parameters.

Estimated values in a spreadsheet model that Crystal Ball defines with a probability distribution.

Describes a relationship or correlation between values of the same data series at different time periods.

Describes a relationship similar to autocorrelation, except instead of the variable being related to other independent variables, it is related to previous values of its own data series.

A relationship between two variables where changes in one independent variable not only correspond to a particular increase or decrease in the dependent variable, but actually cause the increase or decrease.

A statistical summary of the assumptions in a spreadsheet model, output graphically or numerically.

The number of data points minus the number of estimated parameters (coefficients).

In multiple linear regression, a data series or variable that depends on another data series. You should use multiple linear regression as the forecasting method for any dependent variable.

Double exponential smoothing.

Double exponential smoothing applies single exponential smoothing twice, once to the original data and then to the resulting single exponential smoothing data. It is useful where the historic data series is not stationary.

Smooths out past data by performing a moving average on a subset of data that represents a moving average of an original set of data.

Tests for autocorrelation of one time lag.

The difference between the actual data values and the forecasted data values.

Tests the overall significance of the multiple linear regression equation.

The prediction of values of a variables based on known past values of that variable or other related variables. Forecasts can also describe predicted values based on Crystal Ball spreadsheet models and expert judgements.

A regression method that adds one independent variable at a time to the multiple linear regression equation, starting with the independent variable with the greatest significance.

Optimizes the forecasting parameters to minimize the error measure between a set of excluded data and the forecasting values. Predictor does not use the excluded data to calculate the forecasting parameters.

Separates a series into its component parts: seasonality, trend and cycle, and error. This method determines the value of each, projects them forward, and reassembles them to create a forecast.

Considers the effects of seasonality to be multiplicative, that is, growing (or decreasing) over time. This method is similar to the Holt-Winters’ additive method.

A geometric plane that spans more than two dimensions.

In multiple linear regression, the data series or variables that influence the another data series or variable.

A regression method that adds or subtracts one independent variable at a time to or from the multiple linear regression equation.

Defines the offset when comparing a data series with itself. For autocorrelation, this refers to the offset of data that you choose when correlating a data series with itself.

A type of forecasting that optimizes the forecasting parameters to minimize the error measure between the historical data and the fit values, offset by a specified number of periods (lead).

Measures how closely a line matches a set of data. This approach measures the distance of each actual data point from the line, squares each distance, and adds up the squares. The line with the smallest square deviation is the closest fit.

A starting point for the forecast. For a set of data with no trend, this is equivalent to the y-intercept.

An equation with only linear terms. A linear equation has no terms containing variables with exponents or variables multiplied by each other.

A process that models a variable as a function of other first-order explanatory variables. In other words, it approximates the curve with a line, not a curve, which would require higher-order terms involving squares and cubes.

Mean absolute deviation. This is an error statistic that average distance between each pair of actual and fitted data points.

Mean absolute percentage error. This is a relative error measure that uses absolute values to keep the positive and negative errors from cancelling out each other and uses relative errors to let you compare forecast accuracy between time-series methods.

A case of linear regression where one dependent variable is described as a linear function of more than one independent variable.

A forecast obtained with minimal effort based on only the most recent data; e.g., using the last data point to forecast the next period.

Indicates the probability of obtaining an F or t statistic as large as the one calculated for the data.

Tests the significance of a particular independent variable within the existing multiple linear regression equation.

An interactive table in Microsoft Excel. You can move rows and columns and filter PivotTable data.

Coefficient of determination. This statistic indicates the proportion of the dependent variable error that is explained by the regression line.

A process that models a dependent variable as a function of other explanatory (independent) variables.

The difference between the actual data and the predicted data for the dependent variable in multiple linear regression.

Root mean squared error. This is an absolute error measure that squares the deviations to keep the positive and negative deviations from cancelling out each other. This measure also tends to exaggerate large errors, which can help when comparing methods.

Calculates a seasonal index for historical data that does not have a trend. The seasonal adjustment is added to the forecasted level, producing the seasonal additive forecast.

Calculates a seasonal index for historical data that does not have a trend. The seasonal adjustment is multiplied by the forecasted level, producing the seasonal multiplicative forecast.

The change that seasonal factors cause in a data series. For example, if sales increase during the Christmas season and during the summer, the data is seasonal with a six-month period.

Weights past data with exponentially decreasing weights going into the past; that is, the more recent the data value, the greater its weight. This largely overcomes the limitations of moving averages or percentage change methods.

Smooths out past data by averaging the last several periods and projecting that view forward. Predictor automatically calculates the optimal number of periods to be averaged.

A method that solves a set of equations for the coefficients of a regression equation.

Estimates a smooth trend by removing extreme data and reducing data randomness.

Sum of square deviations. The least squares technique for estimating regression coefficients uses this statistic, which measures the error not eliminated by the regression line.

Singular value decomposition.

Tests the significance of the relationship between the dependent variable and any individual independent variable, in the presence of the other independent variables.

A set of values that are ordered in equally spaced intervals of time.

A long-term increase or decrease in time-series data.

In regression, data series are also called variables.

A type of forecasting that optimizes the forecasting parameters to minimize the average error measure between the historical data and the fit values, offset by several different periods (leads).