Loess curve is a statistical technique for plotting a smooth curve through a set of data points in a scattergram.

A scattergram is a plot of various data points in a graph with a predictor variable as its x-axis and a criterion variable as its y-axis.

Loess is a version of a locally weighted scatterplot smoothing technique. Each smoothed value is determined by a quadratic polynomial taking into account the values of data within a particular span of values of the criterion variable, but giving most weight to the central value of the span, less and less weight to more distant values, and zero weight to values outside the span. The span is then moved along the x-axis and a new smoothed value computed. The size of the span is set by a tension factor determining the proportion of the data points to be included in the span.

In its logic, loess is similar to that of a moving average, but each smoothed value is given by a weighted quadratic least squares regression over the span.

Loess is very similar to Lowess, and some authorities treat them as synonyms. Lowess differs in using a linear polynomial for its regression.

See also

• Lowess

External links

• http://www.mathworks.com/access/helpdesk/help/toolbox/curvefit/ch_data7.html
• http://www.itl.nist.gov/div898/handbook/pmd/section1/pmd144.htm
• http://stat.ethz.ch/R-manual/R-patched/library/stats/html/lowess.html