measurement theory in marketing research. measurement what is measurement? assignment of numerals...
TRANSCRIPT
Measurement Theory in Marketing Research
Measurement
What is measurement? Assignment of numerals to objects to represent
quantities of attributes Don’t measure the object -- measure attributes of the
object Don’t measure a person -- measure their weight, height,
social class, GPA, etc. Definition does not suggest how to measure the
attributes
Measurement
Measurement Scales Four types -- NOIR Nominal -- Number is used for identification
purposes Jon Laczniak is number 5 Matt Laczniak is number 9
Numbers reflect nothing -- just used to identify the person
Measurement
Ordinal -- Number is used to reflect order Jon Laczniak is in 7th grade at Ames Middle School; Ethan
Constant is in 3rd Grade Jon is in a “higher” grade How much higher?
Cannot really tell (depends on programs, etc.) No true 0 and differences between grades is not constant
Interval -- Number reflects “intervals” between attributes
Matt Laczniak scored 96 on his soccer skills test; Jon Laczniak scored a 32 Matt scored 64 points higher than Jon! Is he three times as good? Can we really say that someone has “0” soccer skills?
Measurement
Ratio -- Number has an absolute 0 Andy Laczniak is 22 years old; Jon Laczniak is 11
years old Andy is twice as old as Jon Age has a real (and interpretable) 0
Measurement
XO = XT + (ES + ER) XO= Observed score for some construct
XT = True score for some construct
ES = Systematic Error
ER = Random Error
Measurement
Objective in research -- XO = XT When this happens, the measure is valid
If XR= 0; XO = XT + ES Measure is reliable Free of random error
Measurement
Reliability -- instrument measures the same concept every time it is used (XO = XT + ES)
Validity -- instrument measures what it intends to measure (XO = XT)
Given that XO = XT + ES suggests XO is free of random error -- this indicates reliability Reliability is a necessary, but not sufficient
indicator of validity
Measurement
Assessing Reliability (XO = XT + ES ) Consistent responses across time (test/retest reliability) Internally consistent -- all aspects of the measure work
together Multiple measures (of the same concept) are needed
Laczniak Yogurt is: Good/Bad; Favorable/Unfavorable; Positive/Negative
Coefficient alpha = k (mean inter-item correlation)/{1 +[ (k-1) (mean inter-item correlation)]} Here – 3 items are used to measure attitude Need to calculate correlations between each item (3) and then
compute the mean
Measurement
Calculation of coefficient alpha () k (mean inter-item correlation)/1 +[ (k-1) (mean inter-item correlation)]
Where k = number of items used to measure a concept Thus, if one item is used
Mean correlation = 0 Thus,= 0 Single item measures have reliability = 0
Example K = 3 Mean inter-item correlation = .80 = ??
Rules of Thumb = .70 (for new/exploratory measures of concepts) = .85 (for measures that have previously been shown to be reliable)
Measurement
Indicators of Validity (XO = XT) Face validity
Measure “looks” like it should Best to have others (“expert judges” determine this)
Discriminant Validity Measure does not measure some other concept Correlation with measure of other concept is very low
Convergent Validity Measure corresponds to other measures of this concept Correlation with other measures of this concept is high
An Example of Measurement
Ranking versus Rating Rank -- respondent orders the brands according
to their attitude (Ordinal Scale) Rating -- respondent rates each brand on a similar
scale (Interval Scale) ***
An Example of Measurement
Likert Scales -- scales in which respondents indicate their degree of (dis) agreement with statements about the object Generate large number of statements about the
attitude object (e.g, “Professor Laczniak is an exceptional instructor”)
Classify the statements a priori as (un) favorable
An Example of Measurement
Likert Scales (cont’d) Determine a method of scoring (3-point versus 5-point
versus 7-point or more; use a midpoint or not – 4-point) Agree, Neutral, Disagree Strongly Agree, Agree, Neutral, Disagree, Strongly Disagree
Purify the scale by eliminating ambiguous items (through pre-tests) If an item makes the alpha coefficient lower – drop it
Use the scale Has a midpoint (can say if mean response is above/below it) Is interval scaled -- can make mean comparisons Must develop your own norms (midpoints do not always
apply)
An Example of Measurement
Semantic Differential -- uses a series of 7-point gradations with bipolar adjectives that anchor the beginning and end of each scale
Most commonly used -- “My attitude toward the Compaq brand is:”
Good ___:___:___:___:___:___:___ Bad
Positive ___:___:___:___:___:___:___ Negative
Favorable ___:___:___:___:___:___:___ Unfavorable
Measurement
Internal versus External Validity Internal Validity – ability to demonstrate that the
an observed effect is due to the experimental manipulation (Lab Setting)
External Validity – ability to generalize the results of an experiment beyond the experimental subjects (Real World) Internally valid studies are typically not “real” Externally valid studies typically have less controls
Ideally, we follow a lab study with one in the real world