Scribd
Upload
78 views3 pages

Analysis Analysis of Variance One Way Anova

One-way ANOVA compares means between groups and tests the null hypothesis that group means are equal. It requires independent and normally distributed observations within groups, equal variances between groups, and continuous data. One-way ANOVA can analyze stacked or unstacked data. If significant, post-hoc tests like Scheffe, Tukey, Newman-Keuls, and Bonferroni identify which specific group means differ. The ANOVA table splits total variation into between and within group components to calculate the F-test statistic.

Uploaded by

mai
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
78 views3 pages

Analysis Analysis of Variance One Way Anova

One-way ANOVA compares means between groups and tests the null hypothesis that group means are equal. It requires independent and normally distributed observations within groups, equal variances between groups, and continuous data. One-way ANOVA can analyze stacked or unstacked data. If significant, post-hoc tests like Scheffe, Tukey, Newman-Keuls, and Bonferroni identify which specific group means differ. The ANOVA table splits total variation into between and within group components to calculate the F-test statistic.

Uploaded by

mai
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/ 3

One-way ANOVA

The ONE-WAY ANOVA procedure compares means between two or more groups. It is used to compare the
effect of multiple levels (treatments) of a single factor, either discrete or continuous, when there are
multiple observations at each level. The null hypothesis is that the means of the measurement variable are
the same for the different groups of data.

Assumptions
The results can be considered reliable if a) observations within each group are independent random
samples and approximately normally distributed, b) populations variances are equal and c) the data are
continuous. If the assumptions are not met, consider using non-parametric Kruskal-Wallis test.

How To
 If observations for each level are in different columns – run the STATISTICS ->ANALYSIS OF VARIANCE
(ANOVA)->ONE-WAY ANOVA (UNSTACKED) command.
 For stacked data run the STATISTICS->ANALYSIS OF VARIANCE (ANOVA)->ONE-WAY ANOVA (WITH
GROUP VARIABLE ) command, select a RESPONSE variable and a FACTOR variable. Factor variable is a
categorical variable with numeric or text values.
 LE version includes only ONE-WAY ANOVA (UNSTACKED, W/O POST-HOC TESTS) command, and it is
similar to the “ANOVA - Single Factor” command from the Analysis Toolpak package for Microsoft
Excel and does not include post-hoc comparisons.

Data Layout
The data for one-way ANOVA can be arranged in two ways, as shown below.

Samples for each factor level (group) Factor levels are defined by values of
are in different columns the factor variable

Run the “O NE-WAY ANOVA (UNSTACKED )” command. Run the O NE-WAY ANOVA (WITH GROUP VARIABLE ) COMMAND .
Results
Report includes analysis of variance summary table and post-hoc comparisons.

ANALYSIS OF VARIANCE TABLE

The basic idea of ANOVA is to split total variation of the observations into two pieces - the variation within
groups (error variation) and the variation between groups (treatment variation) and then test the
significance of these components contribution to the total variation.

SOURCE OF VARIATION - the source of variation (term in the model).


SS (SUM OF SQUARES) - the sum of squares for the term.
DF (DEGREES OF FREEDOM ) - the number of the degrees of freedom for the corresponding model term.
𝑑𝑓𝑇𝑜𝑡𝑎𝑙 = 𝑁 − 1
𝑑𝑓𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 = 𝐹𝑎𝑐𝑡𝑜𝑟 𝑙𝑒𝑣𝑒𝑙𝑠 (𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠) − 1
𝑑𝑓𝐸𝑟𝑟𝑜𝑟 = 𝑑𝑓𝑇𝑜𝑡𝑎𝑙 − 𝑑𝑓𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡
MS (MEAN SQUARE) - the estimate of the variation accounted for by this term.
𝑀𝑆 = 𝑆𝑆/𝐷𝐹
F - the F-test statistic, under the null hypothesis is distributed as 𝐹𝑑𝑓 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡,𝑑𝑓 𝑒𝑟𝑟𝑜𝑟 .
𝑀𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠
𝐹= 𝑀𝑆𝐸𝑟𝑟𝑜𝑟
P-LEVEL- the significance level of the F-test. If p-level is less than the significance level 𝛼 – the null
hypothesis is rejected, and we can conclude that not all of the group means are equal.

POST-HOC ANALYSIS (MULTIPLE COMPARISON PROCEDURES)

While significant F-test can tell us that the group means are not all equal, we do not know exactly which
means are significantly different from which other ones. With a comparison procedure we compare the
means of each two groups. The SIGNIFICANT column values show if means difference is significant at the 𝛼
alpha level and we should reject the null hypothesis H0.

SCHEFFE CONTRASTS AMONG PAIRS OF MEANS


Scheffe’s test is most popular of the post hoc procedures, the most flexible, and the most conservative.
Scheffe test corrects alpha for all pair-wise comparisons of means. The test statistic is defined as

(𝑀𝑒𝑎𝑛𝑖 − 𝑀𝑒𝑎𝑛𝑗 )2
𝑞=√ .
𝑀𝑆𝑤𝑖𝑡ℎ𝑖𝑛 (1⁄𝑁 + 1⁄𝑁 )
𝑖 𝑗

The test statistic is calculated for each pair of means and the null hypothesis is rejected if 𝑞 is greater than
the critical value 𝑞𝑐𝑟𝑖𝑡 , as previously defined for the original ANOVA analysis

TUKEY TEST FOR DIFFERENCES BETWEEN MEANS


Tukey’s HSD (honestly significant difference) or Tukey A test is based on a studentized range distribution.
𝑀𝑆𝐸
The test statistic is defined as 𝑞 = (𝑀𝑒𝑎𝑛𝑖 − 𝑀𝑒𝑎𝑛𝑗 )√ 𝑁
.

Tukey test requires equal sample sizes per group, but can be adapted to unequal sample sizes as well. The
simplest adaptation uses the harmonic mean of group sizes as N.

TUKEY B OR TUKEY WSD (WHOLLY SIGNIFICANT DIFFERENCE ) TEST

Tukey’s B (WSD) test is also based on a studentized range distribution. Alpha for Tukey B test is the average
of the Newman-Keuls alpha and the Tukey HSD alpha.

NEWMAN-KEULS TEST

The Newman-Keuls test is a stepwise multiple range test, based on a studentized range distribution. The
test statistic is identical to Tukey test statistic but Newman-Keuls test uses different critical values for
different pairs of mean comparisons - the greater the rank difference between pairs of means, the greater
the critical value. The test is more powerful but less conservative than Tukey’s tests.

BONFERRONI TEST FOR DIFFERENCES BETWEEN MEANS


The Bonferroni test is based on the idea to divide the familywise error rate α among tests and test each
individual hypothesis at the statistical significance level of 1/n times what it would be if only one hypothesis
were tested, i.e. at the significance level of α/n.

FISHER LEAST SIGNIFICANT DIFFERENCE (LSD) TEST


The Fisher LSD test is based on the idea that if an omnibus test is conducted and is significant, the null
hypothesis is incorrect. The test statistic is defined as

2 𝑀𝑆𝐸
𝐿𝑆𝐷 = 𝑡𝛼 √ ,
𝑁

where 𝑡𝛼 is the critical value of the t-distribution with the df associated with 𝑀𝑆𝐸, the denominator
of the F statistic.

References
Design and Analysis: A Researcher's Handbook. 3rd edition. Geoffrey Keppel. Englewood Cliffs, NJ: Prentice-
Hall, 1991.

Experimental Design: Procedures for the Behavioral Sciences – 3rd Edition (1995). Roger E. Kirk Pacific
Grove, CA: Brooks/Cole, 1995.

Handbook of Parametric and Nonparametric Statistical Procedures (3rd ed.). Sheskin, David J.. Boca Raton,
FL, 1989.

You might also like