Tuesday 11 May 2010

Madagascan Lemur Regression In New Excel and Adjusted R-squared

Accessing Excel's Awesome "Regression Engine"

Alt-AY2 (AyeAye - a form of Madagascan Lemur)

This brings up the "Data Analysis" listbox with a range of statistical features inside including Random Number Generation and Analysis of Variance.

Type R three times (one to skip Random Number Generation, the second to skip Rank and Percentile and the third to access Regression).

Interpreting Results

The output of Regression is a summary table of "Regression Statistics". Here is a summary of what the statistics mean:

* Multiple R: This is the Coefficient of Simple Correlation. It is the positive square root of the R-squared
* R-squared: Coefficient of Simple Determination. According to wikipedia, it is "the proportion of variability that is accounted for by the statistical model" or "percentage variation in y that is explained by percentage variation in x"
* Adjusted R-squared: R-squared adjusted by the number of predictors (R-squared tends to overestimate the strength of the association, especially when there are >1 independent variables)

What is the relation between coefficient of correlation and the coefficient of determination? One is the square root of the other. So if R-squared is 1%, then the "Multiple R" or "coefficient of correlation" will be 10%.

Standard Error

Standard error of an estimate tells you the accuracy from your prediction.

Troubleshooting

Regression - LINEST() function returns error - usually because ranges are different sizes.

No comments: