Introducing the SPSS CONJOINT Procedure

The Research and Statistical Support group (RSS) of Academic Computing Services (ACS) has recently acquired the "conjoint analysis" procedure from SPSS Inc. We thought it might be useful to review this procedure briefly.

Conjoint analysis is actually a collection of methods which allow one to measure and analyze consumer preferences about attributes of a product or service. Market researchers are frequently interested in finding out those characteristics of a product or service which consumers deem important. While the ideal product would be best, consumers usually make complex tradeoffs which involve price and quality features of the product. SPSS provides three procedures: CONJOINT, ORTHOPLAN, and PLANCARDS. Currently, the CONJOINT procedure is only accessible from a syntax window. ORTHOPLAN and PLANCARDS are available from the main menu bar under DATA – ORTHOGONAL DESIGN – GENERATE, or DISPLAY.

Conjoint analysis is frequently used by market researchers for aiding in the design stage of a product. Researchers can ask questions such as: What products are most and least important to the consumer? , What product attributes are the least and most important to the consumer? And, What is the highest price acceptable for this product. A conjoint analysis asks the consumer to make choices regarding products by trading off qualities or features against one another. With these ratings, the conjoint analysis allows the researcher to determine the relative importance of each attribute including the most and least preferred levels. Additionally, if background demographics are available to the researcher, market segments can be identified to which tailored products can then be developed.

We will use an example to illustrate these procedures. Suppose we wish to develop a statistical package. Several considerations should go into the design of this product. One feature of a statistical package that might be important to consumers would be "ease of use". We might define a low versus high range on ease of use. If this attribute were the only consideration, the choice would be clear. However, some statistical packages which are easier to use may not be as flexible in that they do not offer a wide range of options. Likewise, a statistical package which is more flexible may be more difficult to use from a usability standpoint. From this we define a low versus high dimension on flexibility. Additionally, the cost of the statistical package should be a concern to most consumers. We may define a low versus high range on cost. Suppose our attributes are presented to the consumer as eight alternatives:

	Ease of Use	Flexibility	Price

1.	High	High	$200
2.	High	High	$800
3.	Low	High	$200
4.	Low	High	$800
5.	High	Low	$200
6.	High	Low	$800
7.	Low	Low	$200
8.	Low	Low	$800

Given these alternatives, product 1 is probably the most preferred while product 8 is probably the least preferred. The preferences for the other products will be determined by what is important to each consumer. SPSS CONJOINT uses the "full-concept" approach for conjoint analysis. In this approach, consumers rank alternative products defined by all attributes as in the table above. For a large number of attributes this becomes too time consuming and fatiguing for the consumer. SPSS CONJOINT uses fractional factorial designs, which present an appropriate fraction of the possible alternatives. The ORTHOPLAN procedure generates orthogonal fractional factorial plans. The PLANCARDS procedure enables you to generate a physical profile which can then be sorted by the respondent to arrive at a ranking. The CONJOINT procedure uses the ordinary least-squares method to produce importance ratings of the attributes, part-worth estimates showing preferences for attribute alternatives, and correlations relating predicted rankings from the conjoint model with observed rankings.

For our example we have the following syntax to generate a 2x2x2 orthogonal design; generate the profiles for ranking; and analysis of the ranked data from 8 respondents:

*Generate Orthogonal Design .
SET SEED 10.
ORTHOPLAN
    /FACTORS=usable 'ease of use' ( 1 'Low' 2 'High') flexible 'flexibility of'+
                      ' product' ( 1 'Low' 2 'High') price 'cost of product' ( 200 'Low' 800 'High')
    /REPLACE
    /MINIMUM 8 .
SAVE OUTFILE=’c:\temp\stat.sav’.

*Generate profiles for ranking
PLANCARDS
    /FACTOR=usable flexible price
    /FORMAT BOTH
    /PAGINATE
    /TITLE 'Profile Number CARD' .

*Ranked profiles from respondents
DATA LIST FREE / ID PREF1 to PREF8.
BEGIN DATA
1.00 3.00 5.00 4.00 1.00 2.00 7.00 8.00 6.00
2.00 1.00 2.00 4.00 6.00 7.00 5.00 3.00 8.00
3.00 2.00 1.00 4.00 3.00 6.00 7.00 8.00 5.00
4.00 4.00 1.00 3.00 2.00 5.00 7.00 6.00 8.00
5.00 1.00 2.00 3.00 5.00 6.00 4.00 7.00 8.00
6.00 1.00 4.00 3.00 2.00 7.00 6.00 5.00 8.00
7.00 8.00 2.00 2.00 6.00 7.00 5.00 3.00 1.00
8.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00
9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00
10.00 1.00 5.00 2.00 3.00 4.00 6.00 7.00 8.00
END DATA.

*Conjoint analysis of data
CONJOINT PLAN='c:\temp\stat.sav'
   /DATA=*
   /SEQUENCE=PREF1 to PREF8
   /SUBJECT=ID
   /FACTORS=USABLE FLEXIBLE PRICE
   /PRINT=ALL
   /PLOT=ALL
   /UTILITY='c:\temp\statutil.sav'.
SAVE OUTFILE='statranks.sav'.

The output includes output for each subject as well as output for the group. The output displays the utility scores for each factor level. By adding these values, the total utility of a specific combination can be calculated. For RANK and SEQUENCE data, the relationship is reversed. Low values represent high preference and generate high utilities. High values represent low preference and generate low utilities. The Pearson’s R and Kendaull’s tau statstics indicate how well the model fits the data. These represent the correlations between the observed and estimated preferences. If the model fits well, these values should be very high.