| 1 |
|
| 2 |
See the file INSTALL for installation instructions. |
| 3 |
|
| 4 |
Contents: |
| 5 |
NAME |
| 6 |
SYNOPSIS |
| 7 |
DESCRIPTION |
| 8 |
ALGORITHM |
| 9 |
LIMITATIONS |
| 10 |
EXAMPLES |
| 11 |
METHODS |
| 12 |
SEE ALSO |
| 13 |
AUTHOR |
| 14 |
LICENSE |
| 15 |
DISCLAIMER |
| 16 |
|
| 17 |
NAME |
| 18 |
Statistics::LineFit - Least squares line fit, weighted or unweighted |
| 19 |
|
| 20 |
SYNOPSIS |
| 21 |
use Statistics::LineFit; |
| 22 |
$lineFit = Statistics::LineFit->new(); |
| 23 |
$lineFit->setData (\@xValues, \@yValues) or die "Invalid data"; |
| 24 |
($intercept, $slope) = $lineFit->coefficients(); |
| 25 |
defined $intercept or die "Can't fit line if x values are all equal"; |
| 26 |
$rSquared = $lineFit->rSquared(); |
| 27 |
$meanSquaredError = $lineFit->meanSqError(); |
| 28 |
$durbinWatson = $lineFit->durbinWatson(); |
| 29 |
$sigma = $lineFit->sigma(); |
| 30 |
($tStatIntercept, $tStatSlope) = $lineFit->tStatistics(); |
| 31 |
@predictedYs = $lineFit->predictedYs(); |
| 32 |
@residuals = $lineFit->residuals(); |
| 33 |
(varianceIntercept, $varianceSlope) = $lineFit->varianceOfEstimates(); |
| 34 |
|
| 35 |
DESCRIPTION |
| 36 |
The Statistics::LineFit module does weighted or unweighted least-squares |
| 37 |
line fitting to two-dimensional data (y = a + b * x). (This is also |
| 38 |
called linear regression.) In addition to the slope and y-intercept, the |
| 39 |
module can return the square of the correlation coefficient (R squared), |
| 40 |
the Durbin-Watson statistic, the mean squared error, sigma, the t |
| 41 |
statistics, the variance of the estimates of the slope and y-intercept, |
| 42 |
the predicted y values and the residuals of the y values. (See the |
| 43 |
METHODS section for a description of these statistics.) |
| 44 |
|
| 45 |
The module accepts input data in separate x and y arrays or a single 2-D |
| 46 |
array (an array of arrayrefs). The optional weights are input in a |
| 47 |
separate array. The module can optionally verify that the input data and |
| 48 |
weights are valid numbers. If weights are input, the line fit minimizes |
| 49 |
the weighted sum of the squared errors and the following statistics are |
| 50 |
weighted: the correlation coefficient, the Durbin-Watson statistic, the |
| 51 |
mean squared error, sigma and the t statistics. |
| 52 |
|
| 53 |
The module is state-oriented and caches its results. Once you call the |
| 54 |
setData() method, you can call the other methods in any order or call a |
| 55 |
method several times without invoking redundant calculations. After |
| 56 |
calling setData(), you can modify the input data or weights without |
| 57 |
affecting the module's results. |
| 58 |
|
| 59 |
The decision to use or not use weighting could be made using your a |
| 60 |
priori knowledge of the data or using supplemental data. If the data is |
| 61 |
sparse or contains non-random noise, weighting can degrade the solution. |
| 62 |
Weighting is a good option if some points are suspect or less relevant |
| 63 |
(e.g., older terms in a time series, points that are known to have more |
| 64 |
noise). |
| 65 |
|
| 66 |
ALGORITHM |
| 67 |
The least-square line is the line that minimizes the sum of the squares |
| 68 |
of the y residuals: |
| 69 |
|
| 70 |
Minimize SUM((y[i] - (a + b * x[i])) ** 2) |
| 71 |
|
| 72 |
Setting the parial derivatives of a and b to zero yields a solution that |
| 73 |
can be expressed in terms of the means, variances and covariances of x |
| 74 |
and y: |
| 75 |
|
| 76 |
b = SUM((x[i] - meanX) * (y[i] - meanY)) / SUM((x[i] - meanX) ** 2) |
| 77 |
|
| 78 |
a = meanY - b * meanX |
| 79 |
|
| 80 |
Note that a and b are undefined if all the x values are the same. |
| 81 |
|
| 82 |
If you use weights, each term in the above sums is multiplied by the |
| 83 |
value of the weight for that index. The program normalizes the weights |
| 84 |
(after copying the input values) so that the sum of the weights equals |
| 85 |
the number of points. This minimizes the differences between the |
| 86 |
weighted and unweighted equations. |
| 87 |
|
| 88 |
Statistics::LineFit uses equations that are mathematically equivalent to |
| 89 |
the above equations and computationally more efficient. The module runs |
| 90 |
in O(N) (linear time). |
| 91 |
|
| 92 |
LIMITATIONS |
| 93 |
The regression fails if the input x values are all equal or the only |
| 94 |
unequal x values have zero weights. This is an inherent limit to fitting |
| 95 |
a line of the form y = a + b * x. In this case, the module issues an |
| 96 |
error message and methods that return statistical values will return |
| 97 |
undefined values. You can also use the return value of the regress() |
| 98 |
method to check the status of the regression. |
| 99 |
|
| 100 |
As the sum of the squared deviations of the x values approaches zero, |
| 101 |
the module's results becomes sensitive to the precision of floating |
| 102 |
point operations on the host system. |
| 103 |
|
| 104 |
If the x values are not all the same and the apparent "best fit" line is |
| 105 |
vertical, the module will fit a horizontal line. For example, an input |
| 106 |
of (1, 1), (1, 7), (2, 3), (2, 5) returns a slope of zero, an intercept |
| 107 |
of 4 and an R squared of zero. This is correct behavior because this |
| 108 |
line is the best least-squares fit to the data for the given |
| 109 |
parameterization (y = a + b * x). |
| 110 |
|
| 111 |
On a 32-bit system the results are accurate to about 11 significant |
| 112 |
digits, depending on the input data. Many of the installation tests will |
| 113 |
fail on a system with word lengths of 16 bits or fewer. (You might want |
| 114 |
to upgrade your old 80286 IBM PC.) |
| 115 |
|
| 116 |
EXAMPLES |
| 117 |
Alternate calling sequence: |
| 118 |
use Statistics::LineFit; |
| 119 |
$lineFit = Statistics::LineFit->new(); |
| 120 |
$lineFit->setData(\@x, \@y) or die "Invalid regression data\n"; |
| 121 |
if (defined $lineFit->rSquared() |
| 122 |
and $lineFit->rSquared() > $threshold) |
| 123 |
{ |
| 124 |
($intercept, $slope) = $lineFit->coefficients(); |
| 125 |
print "Slope: $slope Y-intercept: $intercept\n"; |
| 126 |
} |
| 127 |
|
| 128 |
Multiple calls with same object, validate input, suppress error messages: |
| 129 |
use Statistics::LineFit; |
| 130 |
$lineFit = Statistics::LineFit->new(1, 1); |
| 131 |
while (1) { |
| 132 |
@xy = read2Dxy(); # User-supplied subroutine |
| 133 |
$lineFit->setData(\@xy); |
| 134 |
($intercept, $slope) = $lineFit->coefficients(); |
| 135 |
if (defined $intercept) { |
| 136 |
print "Slope: $slope Y-intercept: $intercept\n"; |
| 137 |
} |
| 138 |
} |
| 139 |
|
| 140 |
METHODS |
| 141 |
The module is state-oriented and caches its results. Once you call the |
| 142 |
setData() method, you can call the other methods in any order or call a |
| 143 |
method several times without invoking redundant calculations. |
| 144 |
|
| 145 |
The regression fails if the x values are all the same. In this case, the |
| 146 |
module issues an error message and methods that return statistical |
| 147 |
values will return undefined values. You can also use the return value |
| 148 |
of the regress() method to check the status of the regression. |
| 149 |
|
| 150 |
new() - create a new Statistics::LineFit object |
| 151 |
$lineFit = Statistics::LineFit->new(); |
| 152 |
$lineFit = Statistics::LineFit->new($validate); |
| 153 |
$lineFit = Statistics::LineFit->new($validate, $hush); |
| 154 |
|
| 155 |
$validate = 1 -> Verify input data is numeric (slower execution) |
| 156 |
0 -> Don't verify input data (default, faster execution) |
| 157 |
$hush = 1 -> Suppress error messages |
| 158 |
= 0 -> Enable error messages (default) |
| 159 |
|
| 160 |
coefficients() - Return the slope and y intercept |
| 161 |
($intercept, $slope) = $lineFit->coefficients(); |
| 162 |
|
| 163 |
The returned list is undefined if the regression fails. |
| 164 |
|
| 165 |
durbinWatson() - Return the Durbin-Watson statistic |
| 166 |
$durbinWatson = $lineFit->durbinWatson(); |
| 167 |
|
| 168 |
The Durbin-Watson test is a test for first-order autocorrelation in the |
| 169 |
residuals of a time series regression. The Durbin-Watson statistic has a |
| 170 |
range of 0 to 4; a value of 2 indicates there is no autocorrelation. |
| 171 |
|
| 172 |
The return value is undefined if the regression fails. If weights are |
| 173 |
input, the return value is the weighted Durbin-Watson statistic. |
| 174 |
|
| 175 |
meanSqError() - Return the mean squared error |
| 176 |
$meanSquaredError = $lineFit->meanSqError(); |
| 177 |
|
| 178 |
The return value is undefined if the regression fails. If weights are |
| 179 |
input, the return value is the weighted mean squared error. |
| 180 |
|
| 181 |
predictedYs() - Return the predicted y values |
| 182 |
@predictedYs = $lineFit->predictedYs(); |
| 183 |
|
| 184 |
The returned list is undefined if the regression fails. |
| 185 |
|
| 186 |
regress() - Do the least squares line fit (if not already done) |
| 187 |
$lineFit->regress() or die "Regression failed" |
| 188 |
|
| 189 |
You don't need to call this method because it is invoked by the other |
| 190 |
methods as needed. After you call setData(), you can call regress() at |
| 191 |
any time to get the status of the regression for the current data. |
| 192 |
|
| 193 |
residuals() - Return predicted y values minus input y values |
| 194 |
@residuals = $lineFit->residuals(); |
| 195 |
|
| 196 |
The returned list is undefined if the regression fails. |
| 197 |
|
| 198 |
rSquared() - Return the square of the correlation coefficient |
| 199 |
$rSquared = $lineFit->rSquared(); |
| 200 |
|
| 201 |
R squared, also called the square of the Pearson product-moment |
| 202 |
correlation coefficient, is a measure of goodness-of-fit. It is the |
| 203 |
fraction of the variation in Y that can be attributed to the variation |
| 204 |
in X. A perfect fit will have an R squared of 1; fitting a line to the |
| 205 |
vertices of a regular polygon will yield an R squared of zero. Graphical |
| 206 |
displays of data with an R squared of less than about 0.1 do not show a |
| 207 |
visible linear trend. |
| 208 |
|
| 209 |
The return value is undefined if the regression fails. If weights are |
| 210 |
input, the return value is the weighted correlation coefficient. |
| 211 |
|
| 212 |
setData() - Initialize (x,y) values and optional weights |
| 213 |
$lineFit->setData(\@x, \@y) or die "Invalid regression data"; |
| 214 |
$lineFit->setData(\@x, \@y, \@weights) or die "Invalid regression data"; |
| 215 |
$lineFit->setData(\@xy) or die "Invalid regression data"; |
| 216 |
$lineFit->setData(\@xy, \@weights) or die "Invalid regression data"; |
| 217 |
|
| 218 |
@xy is an array of arrayrefs; x values are $xy[$i][0], y values are |
| 219 |
$xy[$i][1]. (The module does not access any indices greater than |
| 220 |
$xy[$i][1], so the arrayrefs can point to arrays that are longer than |
| 221 |
two elements.) The method identifies the difference between the first |
| 222 |
and fourth calling signatures by examining the first argument. |
| 223 |
|
| 224 |
The optional weights array must be the same length as the data array(s). |
| 225 |
The weights must be non-negative numbers; at least two of the weights |
| 226 |
must be nonzero. Only the relative size of the weights is significant: |
| 227 |
the program normalizes the weights (after copying the input values) so |
| 228 |
that the sum of the weights equals the number of points. If you want to |
| 229 |
do multiple line fits using the same weights, the weights must be passed |
| 230 |
to each call to setData(). |
| 231 |
|
| 232 |
The method will return zero if the array lengths don't match, there are |
| 233 |
less than two data points, any weights are negative or less than two of |
| 234 |
the weights are nonzero. If the new() method was called with validate = |
| 235 |
1, the method will also verify that the data and weights are valid |
| 236 |
numbers. Once you successfully call setData(), the next call to any |
| 237 |
method other than new() or setData() invokes the regression. You can |
| 238 |
modify the input data or weights after calling setData() without |
| 239 |
affecting the module's results. |
| 240 |
|
| 241 |
sigma() - Return the standard error of the estimate |
| 242 |
$sigma = $lineFit->sigma(); |
| 243 |
|
| 244 |
Sigma is an estimate of the homoscedastic standard deviation of the |
| 245 |
error. Sigma is also known as the standard error of the estimate. |
| 246 |
|
| 247 |
The return value is undefined if the regression fails. If weights are |
| 248 |
input, the return value is the weighted standard error. |
| 249 |
|
| 250 |
tStatistics() - Return the t statistics |
| 251 |
(tStatIntercept, $tStatSlope) = $lineFit->tStatistics(); |
| 252 |
|
| 253 |
The t statistic, also called the t ratio or Wald statistic, is used to |
| 254 |
accept or reject a hypothesis using a table of cutoff values computed |
| 255 |
from the t distribution. The t-statistic suggests that the estimated |
| 256 |
value is (reasonable, too small, too large) when the t-statistic is |
| 257 |
(close to zero, large and positive, large and negative). |
| 258 |
|
| 259 |
The returned list is undefined if the regression fails. If weights are |
| 260 |
input, the returned values are the weighted t statistics. |
| 261 |
|
| 262 |
varianceOfEstimates() - Return variances of estimates of intercept, slope |
| 263 |
(varianceIntercept, $varianceSlope) = $lineFit->varianceOfEstimates(); |
| 264 |
|
| 265 |
Assuming the data are noisy or inaccurate, the intercept and slope |
| 266 |
returned by the coefficients() method are only estimates of the true |
| 267 |
intercept and slope. The varianceofEstimate() method returns the |
| 268 |
variances of the estimates of the intercept and slope, respectively. See |
| 269 |
Numerical Recipes in C, section 15.2 (Fitting Data to a Straight Line), |
| 270 |
equation 15.2.9. |
| 271 |
|
| 272 |
The returned list is undefined if the regression fails. If weights are |
| 273 |
input, the returned values are the weighted variances. |
| 274 |
|
| 275 |
SEE ALSO |
| 276 |
Mendenhall, W., and Sincich, T.L., 2003, A Second Course in Statistics: |
| 277 |
Regression Analysis, 6th ed., Prentice Hall. |
| 278 |
Press, W. H., Flannery, B. P., Teukolsky, S. A., Vetterling, W. T., 1992, |
| 279 |
Numerical Recipes in C : The Art of Scientific Computing, 2nd ed., |
| 280 |
Cambridge University Press. |
| 281 |
The man page for perl(1). |
| 282 |
The CPAN modules Statistics::OLS, Statistics::GaussHelmert and |
| 283 |
Statistics::Regression. |
| 284 |
|
| 285 |
Statistics::LineFit is simpler to use than Statistics::GaussHelmert or |
| 286 |
Statistics::Regression. Statistics::LineFit was inspired by and borrows |
| 287 |
some ideas from the venerable Statistics::OLS module. |
| 288 |
|
| 289 |
The significant differences between Statistics::LineFit and |
| 290 |
Statistics::OLS (version 0.07) are: |
| 291 |
|
| 292 |
Statistics::LineFit is more robust. |
| 293 |
Statistics::OLS returns incorrect results for certain input |
| 294 |
datasets. Statistics::OLS does not deep copy its input arrays, which |
| 295 |
can lead to subtle bugs. The Statistics::OLS installation test has |
| 296 |
only one test and does not verify that the regression returns |
| 297 |
correct results. In contrast, Statistics::LineFit has over 200 |
| 298 |
installation tests that use various datasets/calling sequences to |
| 299 |
verify the accuracy of the regression to within 1.0e-10. |
| 300 |
|
| 301 |
Statistics::LineFit is faster. |
| 302 |
For a sequence of calls to new(), setData(\@x, \@y) and regress(), |
| 303 |
Statistics::LineFit is faster than Statistics::OLS by factors of |
| 304 |
2.0, 1.6 and 2.4 for array lengths of 5, 100 and 10000, |
| 305 |
respectively. |
| 306 |
|
| 307 |
Statistics::LineFit can do weighted or unweighted regression. |
| 308 |
Statistics::OLS lacks this option. |
| 309 |
|
| 310 |
Statistics::LineFit has a better interface. |
| 311 |
Once you call the Statistics::LineFit::setData() method, you can |
| 312 |
call the other methods in any order and call methods multiple times |
| 313 |
without invoking redundant calculations. Statistics::LineFit lets |
| 314 |
you enable or disable data verification or error messages. |
| 315 |
|
| 316 |
Statistics::LineFit has better code and documentation. |
| 317 |
The code in Statistics::LineFit is more readable, more object |
| 318 |
oriented and more compliant with Perl coding standards than the code |
| 319 |
in Statistics::OLS. The documentation for Statistics::LineFit is |
| 320 |
more detailed and complete. |
| 321 |
|
| 322 |
AUTHOR |
| 323 |
Richard Anderson, cpan(AT)richardanderson(DOT)org, |
| 324 |
http://www.richardanderson.org |
| 325 |
|
| 326 |
LICENSE |
| 327 |
This program is free software; you can redistribute it and/or modify it |
| 328 |
under the same terms as Perl itself. |
| 329 |
|
| 330 |
The full text of the license can be found in the LICENSE file included |
| 331 |
in the distribution and available in the CPAN listing for |
| 332 |
Statistics::LineFit (see www.cpan.org or search.cpan.org). |
| 333 |
|
| 334 |
DISCLAIMER |
| 335 |
To the maximum extent permitted by applicable law, the author of this |
| 336 |
module disclaims all warranties, either express or implied, including |
| 337 |
but not limited to implied warranties of merchantability and fitness for |
| 338 |
a particular purpose, with regard to the software and the accompanying |
| 339 |
documentation. |
| 340 |
|