• There are fundamentally two reasons for performing a conjoint analysis. The first reason is a measurement reason. If, for example, there are 10 alternatives that could be created from a set of components, having each individual state their preference for each alternative compared with each other alternative would require 45 paired comparisons. Typically, conjoint analysis solves this problem by breaking the choices down into smaller units. The task is spread over many people, and mathematics are used to reconstruct the likely outcome "as if " everyone had been able to test all real alternatives. The problem becomes clearer when we look at a more realistic number of alternatives. If the realistic alternatives could be assembled into 21 "packages" that make sense, that would require 210 comparisons. That would be quite a burden to place on any one individual. Therefore each person evaluates a subset of the components and mathematics assembles the likely "winning" package. Of course, at the end of this process, prudence dictates you should test the winning package to be sure it performs the way the math suggests it will.
• Now, there is Conjoint 2.0, a better way. Using the Linescale, each person can make a direct comparison of each of the 10 alternatives - or even 21 alternatives. Depending on complexity of the alternatives, each individual can rate 11 items on one Linescale and 10 items on another. Or each individual can rate seven complex items on three Linescales. Linescale uniquely creates a head-to-head paired comparison of all items on the Linescale as well as measuring the metric scale values. Give each Linescale a common anchor, and the preference results are immediate and obvious!
• The second reason for using conjoint analysis is to understand the contribution of each component to a whole. This is typically done in order to construct a sensible "package" using the components. With the Linescale approach, each individual can rate the components separately on a single Linescale. Each individual can also separately evaluate the whole, as described above. The combination of these two tasks allows us to create correlation coefficients and the slopes of those correlations which allow us to both (a) determine the winning proposition, and (b) "deconstruct" the winning proposition in order to understand the relative contribution of each of the components.
• The second reason is far less important. Once you understand the winning proposition, and consumers reasons for so rating it, and their verbatim comments, you have a pretty good idea of why it is working as well as what is working.
• In net, Linescale has several advantages over Conjoint 1.0
  • Linescale eliminates the "black box" problem of Conjoint 1.0. You understand "why" the winning proposition wins.
  • The problem of unknown and unpredictable negative interactions of elements is avoided
  • Linescale assumes the client and creatives are better at creating breakthrough propositions than the computer
  • Linescale is fast, with results that are understandable and supportable without resort to "the computer picked it".
  • Preference for the whole proposition is known immediately, with no need for a validating retest. You are ready to immediately go on to product and communication development.

Technical Note / Background

The history of scaling and measurement in supporting better business decisions - and the reasons behind the development of Linescale. For the technically minded…

• Business decision scaling methods over the past 60 years were greatly influenced by work done by Francis Pilgrim and Joe Kamen and others for the U.S. Army Quartermaster Corps during World War ll. Kraft General Foods, Pillsbury, General Mills and many of the other great packaged food companies of the mid-twentieth century - and their advertising agencies - adopted hedonic scale techniques pioneered there. Classic research used nine-point scales, either balanced or unbalanced, to measure food preferences of military personnel. Part of the aim was to measure preferences for categories of foods. For example, were rutabagas preferred to parsnips? Potatoes to rice? Some foods score high; some low. In this classic testing of apples versus oranges, they settled on metric scaling as a convenient, reliable way to measure hedonic norms for categories of foods. These scales were later carried into post-war commercial research for the food industry as packaging and branding of food marketing rolled out.


• Experimentation was done by the more capable research departments, but was generally limited to the number of scale items - nine, seven, six or five items- word choice, balanced or unbalanced scales, graphic representation of the scales, compression and expansion of the psychological space, etc. But with rare exception, such as Eric Marder's constant sum technique, metric scales have been industry practice.


• But scale scores have severe limitations for practical use. Metric scales need large numbers of respondents. Why? They are very "noisy". The noise stems from individual differences. Some people tend to score things high; some score them low. Some score big differences between things, some small. Worse, the same individual will score the same item higher or lower at different times. The net effect is unintended differences that need to be canceled out by large numbers and statistics. Compounding the problem, each product category also carries its own norms; desserts and novelties score high; root vegetables and necessities score low. This forces serious researchers to use large samples. Traditional solutions to this problem include looking at "top box' or "top two boxes". This cancels out much of the scale noise, but also loses information. Canceling out the noise with large samples gets expensive quickly. The dilemma has been forego research, do focus groups or spend a bundle for a large-sample one time study.
  
  
• Why not directly measure preference by ranking? Metric scales are an attempt to infer preference from scale values. An obvious solution is to reframe the problem and go directly to what we really want to measure; relative preference. This is the essence of the Linescale Research technique. Linescale uses sophisticated presentation of scales to derive the simplest and most reliable of measures - rank order preference of items versus the respondents' current favorites. This simple technique solves many traditional measurement problems.


• Rank order is reliable. People may differ over time on how much they like one item or another, but they are quite consistent on their personal order of preference. The net result is reliability and validity with small numbers of respondents. Problems of sampling error never go away, but measurement error (response error) is greatly reduced via the Linescale technique. As a practical and measurable issue, running replications through the data will show thirty to fifty respondents from a properly sampled population are adequate to predict the results of a much larger, more expensive sample. Try it and see.


• Of course you can have your scale metric - more precisely than ever. In effect the Linescale is a 1,000 point scale. Not only do we measure rank order, but we record a finely grained scale score for each individual which we generally report on a six-point basis. Not just "Very Good - but within "Very Good" there are almost 200 independent points of measure. Not just precision of measurement with greater discrimination - but greater reliability than possible with traditional "box" scored scales. Why?


• Reliability comes from anchoring the scale. Each individual anchors the preference Linescale by placing his two competitive favorites on the scale. Thus, the Linescale becomes individually tailored to the individual. The respondent understands the scale, and responds in a familiar context.


• normalization of the scale by each respondent eliminates response error noise, avoids expansion at the scale extremes and decreases variance. A nice package of benefits!

• The Linescore measures the key variables, including whether an experience of the product delivers against the promise. An algorithm calculates the Linescore in real time, and presents it to the respondent. Each person gets to agree or disagree with their score. Then they tell us why they rated it as they did. The ACCEPTORS represent the percent of the sample who scored the tested concept, product or service experience over 600.


• Your ACCEPTOR SCORE is the percent who conclude your website, concept or product is "Excellent, Very strong, Great execution". These are your completely satisfied visitors, members or prospects.