ANOVA or analysis of variance is a topic within statistics which I always get confused with, so this week I have decided to dedicate a blog just for ANOVA, so hopefully it’ll help me understand it by carrying out research and possibly help my readers understand it better if they’re confused as well!
From flipping through a selection of statistics books it appears that ANOVA is used for when you have three or more test conditions.
Basically ANOVA seems to do a similar thing to t-tests, in the sense that the sample needs to have come from a normal distribution. There are other assumptions which are important to consider when using ANOVA, which can often be solved using tests in SPSS, examples of these are Levene’s Test and Greenhouse-Geisser. It is important to consider that if a sample is small when using ANOVA the data maybe skewed, which reduces the chances of the data being normally distributed (which you need for ANOVA).
Analysis of variance can be measured in two ways, either between-groups or within-groups. I am going to look at both to create a better explanation of ANOVA.
Firstly, looking at between-groups variance, as with all ANOVA it is looking at the differences of the means. With between-groups variance specifically it literally wants to look at the difference in the means between the groups. Between-groups occurs in different situations, including treatment effects, individual difference and experimental error.
- Treatment effects–> Is when the differences between means between the groups in big enough to reflect the original manipulation of variables in an experiment.
- Individual differences–> Even though most experiments use random sampling there still can be the problem of individual differences. This can be explained by considering that every person is different and will react differently to experimentation. Despite random sampling you can sometimes have the situation when individuals which are similar to each other end up in one sample group, therefore the results may become unequal across the sample.
- Experimental error–> All experiments may have an element of error, from instructions not properly read out to participants, to more major issues like error in design. However any error in an experiment can affect means and variability.
Secondly looking at within-groups variance, this is simply the variation within in the groups. (Its handy that the names for the different types of variance literally mean what they are called!) Within-group variance can occur from individual difference and experimenter error.
- Individual differences–> Similar to the issue in between-groups, as individuals differ so much, even a random sample may end up not being truly random. Making any scores found unrealistic.
- Experimental error–> Again explained in between-group variance.
A simple explanation to explain the process of ANOVA would be that it tries to find out whether there are differences between the means of the experimental groups (or variables), and it then looks at the differences between each groups individual means against the total mean of all the groups. Essentially what a t-test does, but for more than two experimental groups. It could be argued that if ANOVA does what t-tests do why don’t we carry out multiple t-tests? The answer to this is as follows*, by using many t-tests to test multiple groups can cause complications and can be difficult to follow, therefore using ANOVA should be a quicker and easier alternative.
The total mean of all the groups can also be referred to as the “grand mean”** this is principally the total value for each group divided by the total sample size. The grand mean is useful because it allows you to carry out further analysis of the data, including the total variation. The total variation shouldn’t be confused with variance as it is different; it allows you to compare within-groups to between-groups and vice-versa.
There are many types of ANOVA, but the two key types I will explore are one-way ANOVA and two-way ANOVA. This is the part where I usually get confused, as in I’m never quite sure which type of ANOVA test I should be running, so hopefully this will help!
A one-way ANOVA is defined*** as a statistical test which can only be carried out by numbers, which evaluates the means of two or more groups, when the groups are all using the same single variable. In other words, a test which shows if the mean for two or more groups is different or not.
A two-way ANOVA is defined****as an ANOVA which uses two variables or treatment types. You can also get ANOVA’s which are greater than two-way which use more than two variables.
I think it is also important to be able to read ANOVA on SPSS, as there are many varieties of ANOVA it is obviously important to be able to read all of them, however to prevent my blog being too long winded and for the sake of my TA(!) I’m just going to look at the SPSS output for one-way ANOVA*****.
Like most tests executed on SPSS, the fist table to appear from an ANOVA is of descriptive statistics, which as most of us know basically tells us the means and standard deviations of our data. Following the assumption that one-way ANOVA is testing similar groups, a table showing homogeneity of variances (assuming the test has been carried out to include this!) usually this would be the previously mentioned Levene’s test, which tells you if homogeneity of variance has occurred or not and the Levene’s significance level.
 
And finally what we’ve all been waiting for…and ANOVA table. This tells us if there was any significance in difference between the group means. Significance is marked as “Sig.” and if it is below 0.05, then a statistical significance has been found. The table also gives us useful information about the difference in between-groups and within-groups, degrees of freedom, means square and f-values. ANOVA tables can also be worked out without using SPSS, but that is a story for another day!

Phewww! A lot of information there on ANOVA’s, and I do actually feel as if I understand them better now! A final note before I conclude, I would like to mention that it is sometimes important to carry out post hoc tests after the initial ANOVA has been carried out, this measures the significance difference. And most of the time it is better to use conservative post hoc tests within science.
Right, so to conclude, ANOVA measures basically what a t-test does, but it is specifically designed to test data with more than two test conditions.Â
Â
References
*http://www.experiment-resources.com/anova-test.html
**http://people.richland.edu/james/lecture/m170/ch13-1wy.html
***http://www.google.co.uk/webhp?sourceid=chrome-instant&ix=sea&ie=UTF-8&ion=1#hl=en&q=one+way+anova&tbs=dfn:1&tbo=u&sa=X&ei=mpFoT6CvOZOR8gPF_o35CA&ved=0CB8QkQ4&bav=on.2,or.r_gc.r_pw.r_cp.r_qf.,cf.osb&fp=1e80c3eeca38a663&ix=sea&ion=1&biw=1517&bih=752
****http://www.ilri.org/biometrics/tools/Gloosary.htm
*****https://statistics.laerd.com/spss-tutorials/one-way-anova-using-spss-statistics-2.php
(SPSS examples sourced from above website, with second image edited by myself)Â