
The histogram, when made up of normally distributed data, will form a "bell" curve when a smooth probability density function is produced using kernel smoothing techniques. This line that generalizes the histogram appears to look like a bell.
Often the more data being analyzed and with more resolution will create more bars since more intervals or categories of data are available to plot. The more measurements at various frequencies will create more bars and fill up more of the area under the probability density function.
To assess the data there should be at least 5 bars or intervals and at least 30 data points.
There are a variety of histograms with some explained below. This is also a useful visual tool to depict the skewness and kurtosis of a distribution.
LeftSkewed Distribution (Negatively Skewed):
These histograms
have the curve on the right side or the most common values on the right
side of the distribution. The data extends much farther out to the left
side. These distributions are common where there is an upper
specification limit or it is not possible to exceed an upper value, also
known as boundary limit. This may occur if a customer has requested the
process run at towards the upper specification limit as opposed to
targeting the mean.
The measure of central location is the median.
Mean < Median < Mode
Right Skewed Distribution (Positively Skewed):
The distribution
of the data reaches far out to the right side. This may be caused by a
process having a lower boundary. Cost or time plots commonly exhibit
this behavior.
The measure of central location is the median.
Mode < Median < Mean
If
most common value is 10, the middle most value is 15, and the average
of the data set is 20, then the distribution is right skewed.
Mode = 10
Median = 15
Mean = 20
Bimodal Distribution:
These histograms appear to have two or
more (polymodal) behaviors occurring in one process and appear to have
two points of central location. This can be caused by two sets of data
being analyzed as one that are from different populations such as
plotting the heights of females and males as one distribution.
Uniform Distribution:
The distribution is flat or not
exhibiting much of a bell shape and has no appearance of a central
location. This may occur when all values between a lower specification
limit (LSL) and upper specification limit (USL) are weighted equally
acceptable. In other words, values very close to the limits are as a
good as a value in the middle.
Click here for more information on the Uniform Distribution.
Normal Distribution:
Points are evenly distributed among a central value or location.
The
mean is used to describe the central location of distribution. The
median, mode, and mean are all close to the same value AND the
Coefficient of Skewness is close to zero.
Click here for more information on the Normal Distribution.
The plots below depict a regular and a cumulative histogram of the same data. The data shown is 10,000 points randomly sampled from a normal distribution with mean of 0 and standard deviation of 1. The xaxis labels are represent the zscores.
Notice the cumulative histogram gradually increases to 10,000 (to represent all the data points) and the ordinary histogram shows that data points as they fall into certain data intervals.
Karl Pearson is credited with developing the formula below to measure the Coefficient of Skewness. The formula compares the median with the standard deviation of the same distribution.
If:
Sk > 0 then skewed right distribution
Sk = 0 then normal distribution
Sk < 0 then skewed left distribution
Return to BASIC STATISTICS
Return to the MEASURE phase
Find a career in Six Sigma
Return to the SixSigmaMaterial Home Page
Six Sigma
Six Sigma Modules
The following presentations are available to download.
Green Belt Program (1,000+ Slides)
Basic Statistics
SPC
Process Mapping
Capability Studies
MSA
Cause & Effect Matrix
FMEA
Multivariate Analysis
Central Limit Theorem
Confidence Intervals
Hypothesis Testing
T Tests
1Way Anova Test
ChiSquare Test
Correlation and Regression
SMED
Control Plan
Kaizen
Error Proofing