Scatter Diagrams
Purpose of a Scatter Diagram
Study the relationship between two variables
Find out the extent of variation in one variable due to a
change in the other (eg. Variation in a dimension due to the
change in the lathe machine)
View to control an easy-to-measure characteristic in place of
a characteristic, which is different to measure (eg. Specific
gravity as a measure of concentration)
What is a Scatter Diagram?
Pairs of data of two variables plotted with respect to suitably
marked axes in order to study the relationship between the
variables are called a scatter diagram.
The two variables that are usually dealt with are:
(a) a quality characteristic and a factor affecting it
(b) two related quality characteristics
(c) two factors relating to a single quality characteristic
How to make a Scatter Diagram?
STEP 1
Collect pairs of data STEP 2
(x, y) of the two Find the min. & max. values of
variables & arrange the variables (x & Y).
them in a table. (at Decide the scales of two axis
least 30 pairs data be (the length used to represent
collected) the difference between the min.
& max. for both variable are
approximately equal. (if the
scales are poorly set, it will result in
x-axis (horizontal axis) >> erroneous interpretation of
Independent variable >> factor information.
Y-axis (vertical axis) >>
Dependent variable >> quality characteristic (effect)
How to make a Scatter Diagram? …
STEP 3
Plot the data on the STEP 4
graph paper. Enter all the necessary
When the same values information on the diagram.
are obtained from Especially,
different observations, (a) time interval of data collection
show these points either (b) number of pairs of data
by drawing concentric (c) title and the units of each axis
circles/ plot them in the (d) title of the graph
(e) identity of the person who
immediate vicinity of made the diagram
the first.
Scatter Diagram Example
Scatter Diagram Example …
STEP 1
STEP 2 There are 30 pairs of data
Thickness of the walls (which determines whether a tank is defective or
not) is dependent on the air pressure.
Define x, y variables;
x-axis: air pressure
y-axis: % defective
Maximum & minimum values;
Xmax = 9.4 kgf/cm2 & Xmin = 8.2 kgf/cm2
Ymax = 0.928 & Ymin = 0.864
Mark x, y axis;
x-axis in 0.5 kgf/cm2 intervals from 8.0 to 9.5 kgf/cm2
y-axis in 0.01 % intervals from 0.85 % to 0.93 %
Scatter Diagram Example …
STEP 3
STEP 4
Enter the time interval of the
samples obtained (Oct 1-Nov 9),
number of sample (n=30).
Label the x-axis as blowing air
pressure (kgf/cm2) and y-axis as
percent defective (%.).
State the title of the diagram (i.e.
scatter diagram of blowing air
pressure and per cent defective).
Scatter diagram of blowing air pressure
and percent defective (Oct 1-Nov 9)
Scatter Diagram - Interpretation
Outliers
Outliers are those points, which lie away from the main group of
points.
They may be due to the results of errors in measurement/ recording
data or some change in conditions of operations.
It is necessary to exclude these outliers
before interpreting the diagram.
However instead of neglecting these points
completely, due attention should be paid to
the causes of such irregularities because
often unexpected but very useful information
is revealed during such attempts.
Samples of most common scatter diagrams
Notes on
Scatter Diagrams -
Use proper scales
But this is not so
clear and it
may even
appear to show
no correlation.
If the scales are poorly
selected, the results might give
an erroneous interpretation of
information.
Notes on Scatter keep a
record of
Diagrams - the origin of
data & the
background
Necessary to (useful for
analysis
stratify the data afterwards)
If it is known or suspected that
the origination from different
sources it is advisable to
stratify the data in the scatter
diagram before making any
interpretation.
Data from different strata could
be distinguished in the scatter
diagram by making use of
different colors or symbols.
Notes on Scatter Diagrams -
Correlation depends on the range of the
variables use
There is a
clear positive
correlation
There exists a
correlation greatly
depends on the
range of variables
and it need not
necessarily be the
same for all range.
Testing for CORRELATION using a Scatter
Diagram – MEDIAN METHOD to determined the extent of
correlation without
STEP 1 calculating the coefficient of
Draw the median lines for both axes; correlation
vertical line - equal number of points on
the left and right
horizontal line - equal number of points
above and below the line
Testing for CORRELATION using a Scatter
Diagram – MEDIAN METHOD …
STEP 2
Mark the four quadrants made by the median lines as I, II, III, and IV
starting from the upper right quadrant and going counter clock-wise.
Count the number of points in each quadrant. Denote them by n1, n2, n3,
and n4 respectively.
Testing for CORRELATION using a Scatter
Diagram – MEDIAN METHOD …
STEP 3
Count n1 + n3 = 19 + 20 = 39
Count n2 + n4 = 04 + 05 = 09
Count all points (excluding the points on the line)
N = n1 + n2 + n3 + n4 = 48
If the scatter diagram shws a possible positive correlation, compare the
value of n1 + n3 with the upper limit for the value of N in the sign test table
(at the required confidence limit).
If the value of n1 + n3 is equal or greater than the relevant table upper
limit, there is a definite positive correlation between the two variables.
Testing for CORRELATION using a Scatter
Diagram – MEDIAN METHOD …
STEP 3 …
For the above example,
n1 + n3 = 39 and N = 48
upper limit for N = 48 from the sign test table is 32 (for a confidence level
of 5 %)
Hence n1 + n3 = 39 > 32
Therefore there is a definite positive correlation.
If the scatter diagram shows a possible negative correlation, compare
the value of n2 + n4 with the upper limit for the value of N in the sign test
table. If the value of n2 + n4 is equal or greater than the relevant upper
limit, there is a definite negative correlation between the two variables.
Estimation of the Regression Line
When a scatter diagram shows a linear correlation
between the two variables, the straight line that
describes the relationship between the two variables
is called a regression line.
* It could be used to estimate the value of y for a
given value of x or vice versa.
Estimation of the Regression Line …
STEP 1
Draw the median lines for both
axis of the scatter diagram.
Estimation of the Regression Line …
STEP 2
If the scatter diagram shows a
positive correlation, draw two
median lines (xR and yR) in
quadrant I and two median lines in
quadrant III (xL and yL)
If the diagram shows negative
correlation, median should be
drawn in quadrants II and
IV.
Estimation of the Regression Line …
STEP 3
Join the two points at which the
two medians intersect in each
quadrant.
This is the regression line for the
given set of data.