Computes significance using the F test.
sherpa> FTEST <dof_1> <stat_1> <dof_2> <stat_2>
The command arguments are:
Valid FTEST Command Arguments
The F-test is a model comparison test. Model comparison tests are used to select from two competing models that which best describes a particular dataset.
A model comparison test statistic T is created from the best-fit statistics of each fit; as with all statistics, it is sampled from a probability distribution p(T). The test significance is defined as the integral of p(T) from the observed value of T to infinity. The significance quantifies the probability that one would select the more complex model when in fact the null hypothesis is correct. A standard threshold for selecting the more complex model is significance < 0.05 (the "95% criterion" of statistics).
The F-test may be used if:
If these conditions are fulfilled, then the observed F statistic is sampled from the F distribution, whose shape is a function of <dof_1> and <dof_2>. (The tail integral may be computed analytically using an incomplete beta function; see any basic statistics text for details.) If these conditions are not fulfilled, then the F-test significance may not be accurate.
One can create three F statistics out of the best-fit statistics for two model fits; the most powerful F-test, at least for line detection, uses the change in statistic from one fit to the next as <stat_1>, the additional number of parameters in the more complex model (<delta_dof>) for <dof_1>, the best-fit of the more complex model for <stat_2>, and the number of degrees of freedom for the more complex model for <dof_2>.
The F-test significance can also be retrieved using the Sherpa/S-Lang module function get_ftest.
Compute the F-test significance given two fits:
sherpa> FTEST 2 20.28 34 33.63 significance = 0.000328079
As noted above, for line fits, the first two numbers should be the difference in degrees of freedom and fit statistic, respectively, between the null hypothesis (continuum) fit (for which the chi-square is 53.91 for 36 dof), and the alternative hypothesis (continuum-plus-line) fit (for which the chi-square is 33.63 for 34 dof). In general, if the significance is smaller than a predefined threshold (e.g., 0.05), then the more complex model is selected (as it is in this case). Otherwise the null hypothesis is selected.
The Chandra X-Ray
Center (CXC) is operated for NASA by the Smithsonian Astrophysical Observatory.
60 Garden Street, Cambridge, MA 02138 USA. Email: firstname.lastname@example.org
Smithsonian Institution, Copyright © 1998-2004. All rights reserved.