The StatsChiTest operation supports two types of tests. The first is a comparison of two distributions. The second is a test comparing a binned distribution against expected values.

### Comparing two distributions

In this example we generate two Gaussian pseudo-random distributions with the same standard deviation. We then compute and plot their histograms and finally we run the Chi-Test on the histograms.

Make/o/n=1e5 ddd,eee // create waves

ddd=gnoise(15,2) // fill with Gaussian noise

eee=gnoise(15,2)

Histogram/B={-80,1,160} eee,eh // compute histograms

Histogram/B={-80,1,160} ddd,dh

Graph0() // display histograms

StatsChiTest/T=1 eh,dh

The results in the "Chi-Squared Test" table should be similar to the following:

n | 118 |

df | 117 |

Chi_Squared | 132.712 |

Critical | 143.246 |

P | 0.152077 |

Your result may be somewhat different because the histograms of the waves fluctuate as a result of the pseudo-random number generator. However, because of the way the data were created, the basic Chi-Squared value should be lower than the critical value which is determined by the degrees of freedom and the significance level (set to 0.05 by default). This confirms that the two distributions are the same.

### Comparison with expected values

The second type of Chi-Test is a comparison of binned distribution with expected values. Here we test one of the histograms generated above against an expected distribution contained in the wave called "expected". To run the test execute the commands:

StatsChiTest/T=1/S eh,expected

Graph1()

The results in the "Chi-Squared Test" table should be similar to the following:

n | 124 |

df | 123 |

Chi_Squared | 96.2393 |

Critical | 149.885 |

P | 0.964474 |

In this case again the Chi-Squared value is smaller than the critical value so we can't reject H_{0} that
the tested sample agrees with the expected distribution.

### Sample Size

It is interesting to examine the departure of the histogram from a perfect Gaussian form when we reduce the size of the psuedo-random wave. To test this, execute the following commands:

Make/O/N=10000 fff=gnoise(15,2)

Histogram/B={-80,1,160} fff,fh

StatsChiTest/T=1/S fh,expected2

Graph2()

The results in the "Chi-Squared Test" table should be similar to the following:

n | 104 |

df | 103 |

Chi_Squared | 215.429 |

Critical | 127.689 |

P | 6.06415e-10 |

In this case Chi-Squared is larger than the critical values and the histogram does not match the expected values. Note that expected2 was derived from the wave expected by simple scaling to compensate for the smaller sample size.

Forum

Support

Gallery

### Igor Pro 8

Learn More

### Igor XOP Toolkit

Learn More

### Igor NIDAQ Tools MX

Learn More