1.2. Levels of measurement¶
a.k.a. Scales of measure
Developed by Stanley Smith Stevens
Measurements in science use following scales
Nominal scale
Ordinal scale
Interval scale
Ratio scale
Examples of different scale types
Scale type |
Example |
---|---|
Nominal |
Rocks as igneous, sedimentary, metamorphic; first names |
Ordinal |
Rank-ordering data: Result of horse race, Mohs scale of mineral hardness |
Interval |
temperature with the Celsius scale, |
Ratio |
Mass, length, time, plane angle, energy and electric charge |
Mathematical features of different scale types
Scale type |
Permissible statistics |
Admissible scale transformation |
Mathematical structure |
---|---|---|---|
Nominal |
Mode,Chi-squared |
One-to-one (equality) |
Standard set (unordered) |
Ordinal |
Median,percentile |
Monotonically increasing (order <) |
Totally ordered set |
Interval |
Mean,std,correlation,regression |
Positive linear (affine) |
Affine line |
Ratio |
All above+geometric mean,log etc |
Positive similarities |
1-D vector space |
We can notice, that understanding of scale is important in appropriate choice of statistical measures for data analysis.
1.2.1. Nominal scale¶
Nominal measures offer names or labels for certain characteristics
Objects are classified using labels
Rocks as igneous, sedimentary, metamorphic
A group of people classified based on their first name
Valid operations
Equivalence
Set memberships
1.2.1.1. Categorical variables¶
Variables assessed on a nominal scale are called categorical variables.
Binary variables (or Bernoulli variables) : only two possible categories
Yes or no
success or failure
Multi-way variables
1.2.1.2. Categorical distribution¶
A categorical distribution is a probability distribution that describes the result of a random event that can take one of K possible outcomes, with the probability of each outcome separately specified.
There is not necessarily an underlying ordering of these outcomes.
Numerical labels are attached for convenience.
1.2.1.3. Statistics¶
Central tendency is given by its mode [e.g. most common name]
Mean is not defined [what would be average name in a set of people?]
Median is not defined [There is no ordering in the labels]
1.2.2. Ordinal scale¶
Rank ordering data puts the data on ordinal scale
Order of measurements is described.
Relative size or degree of difference between measured items is not described.
Examples
Result of a horse race, where the horses are ordered based on which one arrived 1st, second, or third, etc..
Names arranged in alphabetical order
We can say which name comes first which later in this order.
But there is no meaning of difference between names.
Psychometric measurements [like IQ etc.]
Food quality : exceptional, great, good, average, bad, poor
1.2.2.1. Statistics¶
Central tendency specified using mean or median
Mean cannot be defined
1.2.2.2. Order isomorphism¶
An ordinal scale defines a total preorder of objects
Scale values may be sorted on a single line with no ambiguities
Numbers may be assigned to scale values
Any transformation of numbers using a monotonically increasing function doesn’t change the order, hence retains validity.
This is known as order isomorphism.
1.2.3. Interval scale¶
Quantitative attributes are measurable on interval scale
Difference between levels is meaningful.
Difference can be multiplied to exceed or equal another difference.
Ratio between numbers of this scale is not meaningful
Multiplication and division cannot be done directly.
Ratio between differences is meaningful.
Choice or origin is arbitrary and not meaningful. [e.g. 0 degree Celsius]
1.2.3.1. Statistics¶
Central tendency can be represented by mode, median, mean all.
Statistical dispersion can be measured using, standard deviation, quantiles etc.
Studentized range or coefficient of variation is not supported.
Moments are not useful since origin is arbitrary. Central moments make sense.
1.2.4. Ratio scale¶
Measurement is the estimation of ratio between magnitude of a continuous quantity and a unit magnitude of same kind.
A zero value is supported.
Examples
Mass, length, time, plane angle, energy and electric charge
Kelvin temperature
1.2.4.1. Statistics¶
Since all mathematical operations are supported, hence all statistical measures are available.
Mode, median, arithmetic mean
Geometric mean, harmonic mean
Range, standard deviation
Studentized range, coefficient of variation
1.2.5. References¶
Change log
- Last Modified
$Id: measurements.rst 249 2012-08-05 06:17:57Z shailesh $