Levels of Measurement
In part, the type of inferential statistics used depends upon the level of
measurement. The four levels of measurement are: nominal, ordinal, interval, and ratio.
Nominal Measurement: This refers to categories of measurement. There is no numerical value. For example, gender, race, and diagnosis are all examples of nominal measurement. This is the weakest form of measurement.
Ordinal Measurement: The attributes can be rank-ordered. However, the distances between the rankings do not have any meaning. A Likert Scale is a good example of ordinal measurement.
|Strongly Disagree (1)||Disagree (2)||Neutral (3)||Agree (4)||Strongly Agree (5)|
|I really like statistics|
Although the rankings go from 1 to 5, these numbers are only used to
represent the order of the response. You can not further analyze it to say that a score of
4 (agree) is twice as strong as a score of 2 (disagree). Therefore, we only use the ranked
levels themselves and do not apply any meaning to the distances between those scores.
Interval Measurement: As in ordinal measurement the attributes are ranked, but in interval measurement the distances between the rankings have meaning and are equal in value. The lack of a true zero point is one thing that characterizes interval measurement. In other words, the zero point is arbitrary. The best example of this is in temperature where a 0 in Celcius is the same as a 32 in Fahrenheit. The zero doesn't indicate the true absence of something. Other examples of interval scores are intelligence measures, and manual muscle testing (where a zero doesn't necessary indicate no contraction in the muscle, we just can't determine it by those means).
Ratio Measurement: The most sophisticated type of
measurement, ratio has all of the characteristics of interval measurement, plus a true
zero point. Range of motion, weight, and length are good examples of this type of
If I have a choice, what level of measurement should I use in data collection? Generally you want to use the most sophisticated level. For example, consider the following example.
What is your age?
___ over 68
This is shown as an ordinal variable. However, it weakens our data analysis, especially if age is an important variable for our study. That is because the ages are broken down into 6 ranges, and individuals within those ranges are clumped together. Using this scale, you would give the same value to a person that was 29 and a person that was 38. Even more questionable, you would give the same value to a person that was 69 and a person that was 89. You can see that by doing this you might lose a lot of information about the influence or importance of age.
A better way to do it:
What is your age? _____
If possible, use age as a ratio level measurement. You can always convert
it to an ordinal variable later on (if you want to). However, once you collect the data as
an ordinal variable, you can't later on convert it to a ratio level. That is because the
subject's real age wasn't written down, but was immediately put into a range.
As a summary:
From: Trochim, W. (2000). The Research Methods Knowledge Base, Atomic Dog Publishing