Scribd
Upload
117 views4 pages

Application of Least Significance Difference

The document discusses the method of least significance difference (LSD) for determining which means differ significantly when ANOVA finds an overall significant effect. LSD calculates the smallest difference between two means that would be considered significant, and compares all pairs of means to see if they differ by more than the LSD value. A worked example analyzes protein content results from 5 laboratories, finds Lab A's mean to be significantly lower than the others based on its difference from Lab C exceeding the LSD. LSD is best used when ANOVA already found significant differences and when comparing adjacent ordered means.

Uploaded by

Boobalan Bala
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
117 views4 pages

Application of Least Significance Difference

The document discusses the method of least significance difference (LSD) for determining which means differ significantly when ANOVA finds an overall significant effect. LSD calculates the smallest difference between two means that would be considered significant, and compares all pairs of means to see if they differ by more than the LSD value. A worked example analyzes protein content results from 5 laboratories, finds Lab A's mean to be significantly lower than the others based on its difference from Lab C exceeding the LSD. LSD is best used when ANOVA already found significant differences and when comparing adjacent ordered means.

Uploaded by

Boobalan Bala
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Application of Least Significance Difference (LSD)

We use the analysis of variance (ANOVA) to determine if there is a significant


effect of a factor. However, if there are more than 2 instances of the factor,
then when a significant difference is found, ANOVA does not tell us which of
those instances contributes significantly to the difference.

For example, suppose we use ANOVA to decide if there is a significant


difference in reported values amongst 5 laboratories (k =5) in the analysis of a
whole milk sample for its protein level with 4 replicates (n’ = 4) each. In this
case, the factor is “laboratory”. In here, the total number of data under
consideration is n = 4 x 5 = 20. The question is: if the ANOVA did conclude
there was a significant difference, was there just one laboratory that was
different from the other four, or were there more?

A simple approach to identify which laboratory or laboratories are different in


their mean results is by the method of least significant difference (LSD). It is
used to compute the smallest difference (i.e., the LSD) between two mean
values as if these means have been the only means to be compared with a
Student’s t-test. We can declare the difference between the two means to be
significantly different when the t-value calculated is larger than the LSD.

Suppose we carry out a hypothesis testing on two reported average results, 𝑥̅1
and 𝑥̅2 with ni replicates, where 𝑥̅1 is the smallest value amongst the means
reported.

The relevant hypothesis is:

Ho : 𝑥̅1 = 𝑥̅ 2 or Ho : 𝑥̅1- 𝑥̅2 = 0,

H1 : 𝑥̅1 - 𝑥̅2 ≠ 0,

1
The Student’s t-test is carried out by equation:

𝑥̅1 −𝑥̅2
𝑡𝑜𝑏𝑠 = 1 1
𝑠𝑝 √( + )
𝑛1 𝑛2

The tobs value should be compared to the t(df: n-k) distribution, and the difference

is significant on the 5% significance level if and only if |tobs| > t (=0.025,df=n-k).


This is the case if and only if

1 1
|𝑥̅1 − 𝑥̅2 | ≥ 𝑡(0.025,𝑛−𝑘) . 𝑠√( + )
𝑛1 𝑛2

The right-hand side of this equation is called the least significant difference – or
the 95% LSD-value – for the difference between these two mean values.

If the numbers of repeats, ni in all the laboratories are the same, i.e. n1=n2 = n’,
then the LSD-value is:

2
𝐿𝑆𝐷0.95 = 𝑡0.025,𝑛−𝑘 . 𝑠√
𝑛′

where s is the standard deviation of the means. This equation is the same for
all pairs of 𝑥̅ ′𝑠.

Now, we can compare differences of the 𝑥̅ ′𝑠 to the LSD-value, and see if there
are significant differences.

Worked example

Five test laboratories took part in an inter-laboratory comparison exercise on


the analysis of a whole milk sample for its protein content (%m/m) with four
replicated results reported by each participant. The results collated are
summarized below, with one-way (one-factor) analysis of variance (ANOVA) as
carried out by the Excel Data Analysis Tools.

2
Trial # Lab A Lab B Lab C Lab D Lab E
1 3.08 3.34 3.21 3.27 3.35
2 3.18 3.25 3.36 3.32 3.38
3 3.04 3.29 3.42 3.38 3.39
4 3.09 3.36 3.05 3.12 3.42

Anova: Single Factor

SUMMARY
Groups Count Sum Average Variance
Lab A 4 12.39 3.098 0.0035
Lab B 4 13.24 3.310 0.0025
Lab C 4 13.04 3.260 0.0274
Lab D 4 13.09 3.273 0.0124
Lab E 4 13.54 3.385 0.0008

ANOVA
Source of
Variation SS df MS F P-value F crit
Between Labs 0.17825 4 0.0445625 4.787 0.0109 3.056
Within Labs 0.13965 15 0.00931

Total 0.3179 19

As the F-value was significantly different from the F critical value at  = 0.05
with 4 and 15 degrees of freedom, it was concluded that there were significant
differences amongst the laboratory results with 95% confidence. To find out
which laboratory’s mean result was different from the rest, the LSD method was
used.

The reported mean results were first re-arranged in ascending order:

Lab A Lab C Lab D Lab B Lab E


3.098 3.260 3.273 3.310 3.385

The standard deviation of the means, s, calculated was 0.1054 and the t-critical
value at (20-5) or 15 degrees of freedom at =0.025 (2-tailed) is 2.131. The
LSD-value calculated is, therefore, LSD = 2.131 x 0.1054 x (2/4) = 0.159.

3
Now, the differences between paired results are found to be:

|Lab A – Lab C| = 0.162;

|Lab C – Lab D| = 0.013;

|Lab D – Lab B| = 0.037;

|Lab B – Lab E| = 0.075.

It is obvious therefore that the difference between Lab A and Lab C is


significantly larger than LSD-value of 0.159, indicating that Lab A’s reported
mean value was at odds with the other participants.

One must take note of some limitations when the LSD method is to be applied,
namely:

1. The LSD is valid only when the F-test under ANOVA confirms that
significant differences between sample means do exist;
2. The LSD value is most robust when we compare the adjacent mean
values, i.e., when the sample means under study are arranged in the
order of magnitude.

You might also like