Comparing Treatment Means

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 15

Comparing Treatment Means

If the ANOVA shows that the differences


between treatments are NOT significant
 Then we accept the null hypothesis Ho:
 A=B=C=D
 There are NO differences between the groups (treatments)
 There is NO NEED to compare the means

 Example in Excel
If the ANOVA shows that the differences
between treatments are significant
 We reject Ho
 There are at least two treatments that are significantly different
 BUT – which ones???
 We need to compare the means
Comparison between treatment means:

 If there are four treatments


 A vs B
 A vs C
 A vs D
 B vs C
 B vs D
 C vs D
 There are 6 possible comparisons one can make
 These are called MULTIPLE COMPARISONS
 We did not plan these comparisons but we did it after we found the differences were significant.
Thus this called POST-HOC comparisons
Least Significant Difference (LSD)
 This is the simplest method.
 We find a value called LSD
 If two means differ by a value >LSD then the means are significantly different. If two
means differ by a value<LSD the means are NOT significantly different
 Arrange the means by the value highest to lowest:

C 65.0 We can denote the


23.8 If LSD = 6.6, means:
D 41.2 Then C>D, B, A C 65.0a
2.8
B 38.4 D=B, D>A D 41.2b
21.8
19.0
B>A B 38.4b
A 19.4 A 19.4c

Means with similar letters are not


significantly different
How to calculate the LSD?

Where:
t = critical value from the t-distribution table
MSw = mean square within, obtained from the results of your
ANOVA test
n = number of scores used to calculate the means
Step 1: Find the t-critical value. The t-critical value for α=0.05, dfw = 36 is 2.028).
Make sure you are using the Within groups DF from your results!

Step 2: Insert the given values, the MSE from your results (I used 26.65 from Within groups on the
above table) the t-distribution value from Step 2 into the least significant difference formula:
LSD = 2.028 √ (26.65 * (2/10)) = 4.68.
Put this number aside for a moment,
Step 3: Calculate ybar1-ybar2 from the results. For this example, we get -8.7.

Step 4: Compare Step 2 and Step 3. If |ybar1-ybar2| ≥ LSD1,2 then you can reject the
null hypothesis that the means are the same (H0:μ1 = μ2). Our value of 8.7 is larger than
4.68 so we can reject the null hypothesis.
Using SAS for ANOVA and Means
Comparisons
DATA COMPARE;
INPUT GROUP$ VALUE;
DATALINES;
A 13
A 23
A 18
A 19
A 24
B 33
B 36
B 41
B 34
B 48
C 67
C 64
C 59
D 39
D 41
D 42
;
PROC ANOVA;
CLASS GROUP;
MODEL VALUE=GROUP;
MEANS GROUP/LSD;
RUN;
The ANOVA Procedure

Dependent Variable: VALUE

Source DF Sum of Mean Square F Value Pr > F


Squares

Model 3 5246.800000 1748.933333 72.27 <.0001

Error 16 387.200000 24.200000    

Corrected 19 5634.000000      
Total

R-Square Coeff Var Root MSE VALUE Mean


0.931274 11.99841 4.919350 41.00000

Source DF Anova SS Mean Square F Value Pr > F


GROUP 3 5246.800000 1748.933333 72.27 <.0001
The ANOVA Procedure
t Tests (LSD) for VALUE
Alpha 0.05
Error Degrees of Freedom 16
Error Mean Square 24.2
Critical Value of t 2.11991
Least Significant Difference 6.5956

Means with the same letter


are not significantly different.
t Grouping Mean N GROUP
a 65.000 5 C
       
b 41.200 5 D
     
b 38.400 5 B
       
c 19.400 5 A
Problems with LSD

 Only one value to compare many pairs


 Chances of making Type 1 Error increases with each pair of comparison
 Instead of P<0.05 it can become P<0.35
 Can only be used if there are small number of treatments <5
 To overcome this problem other measures are introduced
 E.g. Duncan’s Multiple Range Test, Tukey’s HSD, Dunnett’s etc.
In SAS many methods can be used

PROC ANOVA;
CLASS GROUP;
MODEL VALUE=GROUP;
MEANS GROUP/LSD, DUNCAN, TUKEYS;
RUN;
END

You might also like