|Module 3: Interpreting Data |
4. Using data to support an argument; making inferences
4.3 What is the statistical meaning of 'significant'?
It is essential that we approach and solve problems by using and interpreting data, not by giving "obvious" answers.
For example, it is commonly believed that older adults have pet animals to compensate for loneliness after losing their partner or their children move away from home. Consider the article Pet ownership linked to Depression  (Horen 2004) for surprising research! Rather than being good for older residents, the researcher found that older Australians who own a pet are more likely to be
- in poorer physical health;
- aggressive and hostile towards the world than people who don't own pets.
So the commonly held hypothesis has been disproved by research: "Flying in the face of claims from the pet food industry, and others, the study shows pet ownership confers no health benefits to older people".
"Significant" in a statistical sense means "not likely to happen just by chance". It does not mean "important" (Moore 1995). Statistical significance is usually expressed in terms of a significance level which is a percentage, but no matter what that percentage is, significance does not equate to importance. In a study in which the sample size is small, an important effect might not show up as being statistically significant. Conversely, an effect that is not at all important might be statistically significant in a large study. Furthermore, the significance of an effect (whether or not an important effect) can be increased by increasing the sample size of the study or through a more effective design.
Statistical significance implies that there is evidence of an association between two variables. However, such analysis does not mean that the alternative hypothesis is true and statistical significance cannot be used to make claims about whether one variable causes changes to another variable.