Before deciding on the measure of central tendency, it is important to examine the frequency distribution of your data. This is because the best measure of central tendency depends on the distribution, as well as the type of data.
In a normal distribution, the frequency distribution is symmetrical and appears like a bell. It is also called a bell curve. However, not all the data is normally distributed and this impacts the accuracy of certain types of central tendency.
The aspects that impact the frequency distribution include skew and kurtosis.
Skew is the lack of symmetry in a distribution, where a cluster of scores appears on one side or the other. If scores are clustered towards the lower end, and the tail points to the higher end, this is a positive skew. If scores are clustered towards the higher end, and the tail points to the lower end, this is a negative skew. The image to the left is an example of a positively and negatively skewed distribution.
Kurtosis is the pointyness of the frequency distribution. If the distribution has heavy tails and is pointy, it has a positive kurtosis and called leptokurtic. If the tails are thin and it is relatively flat, it has a negative kurtosis and called platykurtic. Examples are to the right.
Distributions should have no skew or excess kurtosis. The further from zero the skewness or excess kurtosis is, the further the frequency distribution is from zero.
There are statistical tests, such as the Kolmogorov-Smirnov test and the Shapiro-Wilk test, that determine whether the skewness or kurtosis is significantly different from zero. If a frequency distribution is found to have significant skew or kurtosis, that needs to be considered when choosing a measure of central tendency.
Click here for a link about testing for normality.