Some Approximations Are Better Than Others

Next: Using the TI-83: Up: Computing Binomial Probabilities Using Previous: Computing Binomial Probabilities Using

Some Approximations Are Better Than Others

Examine the shape of the binomial histogram for n=20 and p=.10 in Figure 4.9.

**Figure 4.9:** Histogram of Probabilities for Binomial with n=20 and p=.10
$\begin{figure} \begin{center} \epsfig{file=Bhist201.ps, height=4in, width=3in, angle=-90}\end{center}\end{figure}$

Since there is a noticeable skewness to the right, we will get a poor approximation of areas if we replace the rectangles by a normal curve. If the value of p were .90 instead of .10, the binomial histogram would be skewed left. This is typical behavior of binomial histograms whenever p is either too close to 0 or too close to 1. When is it 'safe' to use the normal curve to approximate binomial probabilities? A convenient rule of thumb is as follows:

$\fbox{ \parbox{5.5in}{ \vspace*{1ex} The normal curve with gives reasonably good... ...omial probabilites whenever both $np>5$\space and $n(1-p)>5$ . \vspace*{1ex} } }$

The reader is reminded that normal curve approximations, no matter how close, are still approximations. The binomial formula ( 4.3) should be used to calculate exact probabilities whenever possible.

2003-09-08