Scaling techniques in Machine Learning

Definition: Scaling is a technique of generating an endless sequence of values, upon which the measured objects are placed. Several scaling techniques are employed to review the connection between the objects.

Following are the two categories under scaling techniques:

Comparative scales:  

It involves the direct comparison of objects. Comparative scale data must be interpreted in corresponding terms and have either ordinal or rank order properties.

Types of comparative scales are:

1. Paired comparison:

• This technique is a widely used comparative scaling technique.
• In this technique, the respondent is asked to pick one object among the two objects with the help of some criterion.
• The respondent makes a series of judgements between objects.
• The data obtained is ordinal in nature.
• With n brands, [n(n-1)/2] paired comparisons are required.

For example: A survey was conducted to find out consumer's preference for dark chocolate or white chocolate. The outcome was as follows:

Dark chocolate= 30%
White chocolate= 70%

Thus, it is visible that consumers prefer white chocolate over dark chocolate.

2. Rank order:

• In this technique, the respondent judges one item against others.
• Respondent are present with several objects and are asked to rank or order them according to some criterion.
• Rank order scaling is also ordinal in nature.
• Only (n-1) scaling decisions need to be made in this technique.

For example: A respondent is asked to rate the following soft drinks:

3. Constant sum scaling:

• In this technique, the respondent is assigned with the constant sum of units, such as 100 points to attributes of a product to reflect their importance.
• If the attribute is not important, the respondent assigns it 0 or no points.
• If an attribute is twice as important as another attribute, it receives twice as many points.
• The sum of all points is 100, that is, constant. Hence, the name of the scale.

For example: A respondent is asked to spend 500 rupees on product A, product B and product C of foods?

4. Q sort:

• It is a sophisticated form of rank order.
• In this technique, a set of objects is given to an individual to sort into piles to specified rating categories.

For example: A respondent is given 10 brands of shampoos and asked to place them in 2 piles, ranging from "most preferred" to "least preferred".

pile 1

                                               Most preferred

pile 2

                                               Least preferred

Note: Generally the most preferred shampoo is placed on the top while the least preferred at the bottom.

Non-comparative scales:

 In non-comparative scales, each object of the stimulus set is scaled independently of the others. The resulting data are generally assumed to be ratio scaled.

Types of Non-comparative scales are:

1. Continuous rating scales: 

• It is a graphic continuum typically coordinated by two extremes.
• The extreme values are not predefined.
• It can be constructed easily and is simple to use.
• The respondent rates by placing the mark on a continuous line.

For example: A respondent is asked to rate the service of Domino's:

Type 1

Following are the two categories under scaling techniques:

Type 2

2.  Itemised rating scales: 

• It is a graphic continuum typically coordinated by two extremes.
• It is simple to use and can be constructed easily.
• The respondent is provided with a scale that has a number or brief description associated with each category.
• The categories are ordered in terms of scale position, and therefore the respondents are required to pick the required category that best describes the object being rated.

The different forms of Itemised rating scales are - a. Itemised graphic scale, b. Itemised verbal scale, c. Itemised numeric scale

Types of Itemised rating scaling are:

a. Likert scale: 

• This scale requires the respondent to indicate a degree of agreement or disagreement with the statements mentions on the left side of the object.
• The analysis is often conducted on an item-by-item basis, or a total score can be calculated.
• When arriving at a total score, the categories assigned to the negative statements by the respondent is scored by reversing the scale.

For example: A well-known shampoo brand carried out Likert scaling technique to find the agreement or disagreement for ayurvedic shampoo.

b.  Semantic differential scale:

• The semantic differential is a 7 point rating scale with endpoints related to bipolar labels.
• The negative words or phrase sometimes appears on the left side or sometimes right side.
• This controls the tendency of the respondents, particularly those with very positive and very negative attitudes, to mark the right or left sides without reading the labels.
• Individual items on a semantic differential scale could also be scored on either a -3 to +3 or 1 to 7 scale.

For example:  A well-known shoe brand carried out semantic differential scaling technique to find out customer's opinion towards their product.

c. Staple scale:

• It is a unipolar rating scale with 10 categories scaled from -5 to +5.
• It does not have a neutral point, that is, zero.
• It is represented vertically.

For example: A well-known shoe brand carried out a staple scaling technique to find out costumer's opinion towards their product.