Measurement & Scaling (Research)

Measurement & Scaling in research / business (Research)

Measurement & Scaling in research / business research.

Measurement & Scaling: Introduction

Measurement

Assigning numbers or other symbols to characteristics of objects according to certain specified rules.

• Numbers permit statistical analysis of data
• Numbers facilitate the communication of measurement of rules and results

Specification of rules for assigning numbers to the characteristics is especially important
Assignment process must be isomorphic , one to one correspondence between the number and characteristics being measured
Rules should be standardized and applied uniformly

Scaling

Scaling is the generation of a continuum upon which measured objects are located.

Scales of Measurement are: Nominal, Ordinal, Interval, Ratio

Scales of Measurement

Nominal Scale

• Figurative labeling
• Numbers serve as labels/ tags
• Aadhar numbers/ T-shirt numbers of players

Used for identifying respondents, brands, attributes, stores and other objects.
When used for classification purposes, the nominal scaled numbers serve as labels for categories / class
Operation : Counting
Statistics : based on frequency counts

Ordinal Scale

It is a Ranking Scale.
Possible to determine whether an object has more or less of a characteristic than some other objects
Indicates relative position and not the magnitude of difference between the objects
Used to measure relative attitudes, opinions, preferences and perceptions
Counting Operations & operations based on Centiles

Interval Scale

In an interval scale, numerically equal distances on the scale present equal values on the characteristics being measured.
Interval scales contains all the information of an ordinal scale but it also allows to compare the differences between objects.

Operations in addition : Arithmetic mean.

Ratio Scale

All properties of Nominal, Ordinal & Interval scale plus an absolute zero point.
Identify, classify objects , compare intervals and differences.
Can be subjected to all statistical manipulations.

Comparative & Non-Comparative Scale

Scales can be:

Comparative
It involves the direct comparison of stimulus objects. The data must be interpreted in relative terms and have only ordinal properties. Example: Paired Comparison, Rank Order, Q-Sort, Constant Sum

Non-Comparative
Each object is scaled independently of the others in the stimulus set. The data is assumed to be interval or ratio scaled. Examples: Continuous, Itemized Rating Scales, Likert, Semantic Difference, Stapel

Advantages of Comparative Scales

• Small differences can be deducted between stimulus objects
• Same reference points for all respondents
• Easily understood
• Involves fewer theoretical assumptions
• Reduces the halo or carryover effects

Relative Disadvantages of Comparative Scales

• Ordinal Data
• Inability to generalize beyond the stimulus objects scaled

Comparative Scale

Paired Comparison Scaling

A respondent is presented with 2 objects and asked to select one according to some criterion. Data obtained is ordinal. Most widely used comparative technique. For n brands, [ n(n-1)/2] paired comparisons are required

Rank Order Rating

Respondents are asked to arrange a number of objects as per some criterion – taste/ looks/ quality etc.
The result is an ordinal scale which tells you which the most preferred to the least preferred but nothing about the distance between any of the objects.
Most realistic in representing the actual shopping situation.

Rank Order Scaling:

Please rank the following teas on the basis of its refreshing ability ( place 1 besides the rank you think is the most refreshing and 5 for the least refreshing)

Brooke Bond Red Label ( 3)
Tata Tetley ( 2)
Lipton Green Tea(1)
Taj Mahal Classc ( 4)
Wagh Bakri CTC ( 5)

Constant Sum

Respondents are to allocate a given number of points among the given objects according to some criterion.

Respondents are advised to allocate their points in proportion to their preference for the object.

Please divide the following 100 points among the following cars on the basis of fuel efficiency.

Maruti Alto – 50
Chevorlet Spark -20
Honda Brio – 30

Non-Comparative Scale

• Continuous Scale
• Interval Scale: Itemized Rating Scale, Semantic Differential Scale, Stapel Scale

Continuous Rating Scale

Graphic Rating Scale
Scale runs from one extreme of the criterion variable to another extreme
Respondents mark an appropriate position on the scale

Itemized Rating Scale

Likert Scale

Measurement scale with 5 response categories ranging from strongly disagree to strongly agree.

The respondent is to indicate a degree of agreement /disagreement with each of a series of statement.

Example of Likert Scale:

Statements: SD = 1, Disagree =2, Neither A nor D = 3, Agree = 4, S A =5
Flipkart offers a good mix of brands (Enter the choice)
Flipkart offers reasonable prices (Enter the choice)
Flipkart customer service is good (Enter the choice)

Semantic Differential Scale

7 point rating, Endpoints have bipolar labels.

For example:

Tata automobiles are:
Reliable _ X_ _ _ _ _ _ _ _ Unreliable
Modern _ _ _ _ _X _ _ _ _ Old – Fashioned

Staple

Measures attitude that consist of single adjective in middle of an even numbered range of values from -5 to + 5 without a neutral point.

Commonly used Scales

Importance

Very Important, Important, Neither Important nor Unimportant, Unimportant, Very Unimportant

Satisfaction

Very Satisfied, Satisfied, Neither Satisfied nor Unsatisfied, Unsatisfied , Very Unsatisfied

Likely

Highly Likely, Likely ,Unlikely, Highly Unlikely

Scale Evaluation

• Error
• Reliability: Test/ Retest
• Validity
• Generalizability

Measurement Error

Variation on the information sought by researchers and the information generated by the measurement process employed.
Two types of Errors: Systematic Errors, Random Error.

Reliability

Consistent Results on repeated measurement
Impact of Systematic error on Reliability
Impact of Random error on Reliability

Validity

Extent to which difference in observed scale scores reflect true differences among objects on the characteristic being measured rather than Systematic Error or Random Error. (Perfect Validity would then be ?).

Relationship between Reliability and Validity:

Perfectly Valid = Perfectly Reliable
Reliable “is not equal to” Perfectly Valid