Here are some thoughts from me on metric and nonmetric data, does it match yours? What would you add, change, query?
Metric data
Metric data is what most people mean when they talk about ‘numbers’, the sorts of numbers we collect when we measure something. How old somebody is an example of metric data, so is a rating scale where somebody indicates how likely they are to recommend a product or service on a 0 to 10 scale.
Metric data can be subdivided into sub-categories and two key ones are ratio scales and interval scales.
Ratio Scale
Ratio scale is the term we use where there is a ‘real’ zero, i.e. a 0 means a total absence of whatever is being measured. For example, ‘How many times did you eat curry this month?’ produces ratio scale data. Somebody who ate curry 8 times in the last month ate it twice as many times as somebody who had curry 4 times. Only ratio scales allow us to say that 8 is twice as big as 4, or that the increase from 4 to 6 is an increase of 50%.
Interval Scale
Interval scale describes a scale where the units of the scale measure a consistent concept. In an interval scale the gap between 2 and 4 is the same as the gap between 6 and 8. However, an interval sale does not have a ‘real’ zero.
Both Fahrenheit and Celsius are examples of interval scales. If the temperature moves from 50°F to 70°F, the change is 20°F; if the temperature changes from 80°F to 100°F that is also a change of 20°F. However, 100°F is not twice as hot as 50°F, because 0°F does not mean there is no heat at all.
In market research many of the scales we use, such as the five-point Agree-Disagree scale produce interval scale data. For example, if the scale uses 1 for Disagree Strongly through to 5 for Agree Strongly, then a change of the mean from 2 to 3 is not a 50% increase, and a mean of 4 is not twice as big as a mean of 2.
Integer Scale
Integer scale refers to a metric scale (interval or ratio) where only whole numbers (numbers without decimal values) are used. The number people in a classroom is an integer ratio scale, there are no fractions of people. A five-point Agree-Disagree scale is an integer interval scale.
Cardinal Numbers
The term cardinal is often used in comparison to ordinal (where ordinal refers to ranked data). Cardinal numbers refer to counting things, such as number of children, number of days etc. Cardinal numbers are either 0 or a positive integer. In the context of market research, the term cardinal should probably be avoided; terms like interval or ratio scale are usually more helpful.
Why does it matter?
Many of the statistical techniques we use in market research, including the mean and standard deviation only ‘work’ with metric data. The term ‘work’ in this context expresses the probability we will get reliable and meaningful answers if we use these techniques.
Nonmetric Data
Nonmetric data refers to all the structured data market researchers use that is not metric data. For example, nonmetric data includes information that is ranked (which is called ordinal) and information that has no linear pattern to it (which is called nominal or categorical).
Ordinal Data
An example of ordinal data would be a question where people were shown four holidays and asked to rank them from first choice through to last choice. Another name for ordinal data is ranked data.
Sometimes ordinal data is converted to metric data by transforming the data. The Formula One Car Championship shows one example of this transformation process. In 2017, points are awarded for cars that finish in the first ten places for each race, the winner (i.e. 1st) is awarded 25 points, 2nd awarded 18 points, 3rd is given 15 points, down to the 10th car receiving 1 point. These points are accumulated over the season and create an integer scale.
Nominal or Categorical Data
A simple nominal scale might be “Are you? Male or Female?”, a slightly more complex one would be something like “Which region do you live in? Asia, Australasia, Europe, North America, South America, Other”.
In market research categorical data can be in the form of mutually exclusive categories, such as “Which of the following would be your first choice? Beef, Pork Chicken, None of these”. However, it can also permit multiple selections, such as “Which of the following do you eat at least once a month? Beef, Pork, Chicken, None of these”.
The decision about whether some scales are categorical or ordinal can sometimes be quite subjective. For example, the question “What was the highest level of education you completed?” followed by a series of options from specific years in High School through to post-graduate levels, could be considered categorical (i.e. purely descriptive) or it could be considered ordinal (with “Post Doctoral Studies” being at one end of the ranking, and perhaps “Some High School” being the other end.
Dichotomous Data
Categorical data with just two options (for example a gender question with just the options Male and Female) can also be treated as a special type of variable, referred to as a dichotomous variable. This sort of variable can be coded as a 1 and 0 (for example a 1 if somebody is Female, and 0 if they are Male), which makes the variable suitable for a range of statistical techniques such as logistical regression. Choice data (as opposed to ranking or rating) tends to produce dichotomous data and the analysis techniques tend to be based on techniques based on all the data being 1 or 0. This is a growth area in terms of analysis in market research.
Why does it matter?
With non-metric data we can’t use tools like mean, standard deviation, z tests, or factor analysis. We use a family of statistics called non-parametric statistics, such as counts, percentages, chi-square, mode and median. In the case of dichotomous variables we also have the opportunity to use logistic techniques.