Download as pdf or txt
Download as pdf or txt
You are on page 1of 1

Control Charts for Symmetrical

Distribution

Data with Positive Kurtosis

its use. T h e c o m p u t a t i o n of t h e limits of t h e "classical" control charts assumes t h a t t h e distribution of the d a t a must be normally distributed. If this is not t h e case, t h e t y p e I error a really obtained will be different t h a n the expected t y p e I error a = 0.0027 (3a limits). But how large is t h e difference between the observed and t h e expected type I error when t h e kurtosis increases? In order t o evaluate the impact of t h e kurtosis 72 of t h e d a t a on the observed type I error we used t h e following approach: for a sample size n G {5, 7, 9}. for a kurtosis 72 [0,10], mean m i = 0, and variance \i2 = \. compute b and d using Eqs. (3) and (4). generate (by inverse simulation of Eq. (2)) a set of m samples of n symmetrical Johnson Su r a n d o m variables having parameters (b, d). T h e number m of sample has been chosen such t h a t the total number of generated d a t a isTOx ro = 3 465 000. compute t h e mean, median, s t a n d a r d deviation, and range for each sample. compute t h e proportion of d a t a outside the control limits for each of t h e four statistics. For a kurtosis 72 0, the estimated observed proportion of d a t a outside the control limits (the t y p e I error) must be close to the expected one a = 0.0027, for all the control charts.

In Fig. 3 we have plotted the observed t y p e I error versus t h e kurtosis for sample size n = 5, 7,9. T h e conclusions of these simulations are: T h e mean and median charts seem t o be very insensitive to t h e kurtosis of t h e data. As expected, this is particulary t r u e for the median. T h e larger t h e sample size, the more insensitive the charts. In contrary, t h e s t a n d a r d deviation and range charts seem t o be very sensitive t o the kurtosis. T h e range chart is t h e most sensitive. T h e larger t h e sample size, the more sensitive the charts. From this, we can conclude t h a t t h e m e t h o d proposed in this paper is mostly devoted to t h e computation of t h e limits of s t a n d a r d deviation or range charts (dispersion charts in general) of d a t a having a positive kurtosis, even if it can also be applied t o t h e other statistics. 6. O C a n d A R L C u r v e s Let X be a symmetrical Johnson Su with parameters (b,d). W i t h o u t loss of generality, we will assume t h a t TOI = 0. By definition, the random variable Z defined by:

Z = 6sinh-1^j

(6)

is a normal (0,1) r a n d o m variable. In order to compute the OC and ARL curves of a control chart using the method proposed in this paper, we have to first find t h e distributions of t h e random variables U and V defined as:

You might also like