calibri font

Residential customer needs Appendix 6 – Conjoint analysis

1 of 21

Appendix 6 – Conjoint analysis

This appendix provides a short explanation of conjoint analysis and further

details about the propositions tested and the way the results have been

analysed

Conjoint analysis

Conjoint analysis is a statistical technique that helps us better understand

what people really value in products and services and how they make these

decisions. Every customer making choices between products and services is

faced with trade-offs. Is high quality more important than a low price and quick

delivery for instance? Customers find it difficult to answer these questions

directly and rationalise their choices, partly because we are asking them to

think about their preferences in a way that is unfamiliar to them, and partly

because other factors such as the desire to appear logical or socially

responsible constrain their responses.

Conjoint enables us to obtain the information on what choices people make

and what is driving their behaviour in a simpler and more reliable way. In a

conjoint exercise respondents are asked to choose between different product

concepts - descriptions of the full product or service with different

combinations of the component features. The task is therefore much more

straightforward for the respondent than asking them to rationalise the choices

they make themselves, as we simply ask what they would choose.

Conjoint analysis is used to analyse the choices people made and to help us

understand which features are driving those choices. The analysis provides

us with scores that not only summarise the influence different features have,

but that can also be used to model the appeal or acceptability of any

combination of the features tested, not just those combinations which were

evaluated in the exercise. Conjoint therefore provides us with an identification

of what is really driving customer behaviour, rather than what customers say

is important to them, and also offers a means of evaluating many more

product combinations than we could reasonably ask a respondent to do

directly.


2 of 21

Propositions tested

The following figure summarises the propositions tested. Each feature is

represented by an attribute that has between two and four levels which define

the range of different levels of service which may be available within that

feature. The levels include the current service level.

Figure 1 - Summary of propositions Attribute Levels

75% of first class post arrives within 1 day

85% of first class post arrives within 1 day

First class quality of service

93% of first class post arrives within 1 day*

Single Service Single service

95% of post arrives within two days

Cost = 35p

No facility to specify evening / Saturday

delivery

Ability to specify evening /

Saturday delivery of items

requiring a signature or too

large to fit through letterbox Sender can specify an evening or Saturday

delivery for an additional fee of £1.00 per

item

Single insured & next day guaranteed

delivery service (£5.05)

Insured/guaranteed items

Insured service (£2.95)

Next day guaranteed delivery (£1.95)

5 days a week (Monday to Friday)

5 days a week (Saturday and 4 weekdays)

Collection and delivery of mail

6 days a week (Monday to Saturday)

first class = 41p

second class = 32p

first class = 45p

second class = 35p

first class = 49p

second class = 38p

Cost of postage

first class = 55p

second class = 42p

*The quality of service for second class was always shown at 98% (the standard is 98.5%,

but was rounded down in the questioning)


3 of 21

Methodology

Rather than ask about attributes independently, the conjoint requires

respondents to assess them as part of a package. This avoids the pitfalls of a

simple evaluation of individual attributes which can often result in everything

being important and little real discrimination between features.

The use of levels within features is also important in aiding more accurate

measurement. Without levels we would simply be asking about the

importance of each attribute as a concept, which may be very easy to answer

but is very difficult to answer accurately or in a way we can really use the

responses. For example, the impact that price has on people’s perceptions of

the service depends on what price range we are considering – if we were put

up the price by 100%, then it is reasonable to assume that this would matter a

lot more to customers than a 10% price rise. It is impossible, therefore, to

have a truly meaningful response to the importance of price without knowing

what the parameters are – so by defining levels within the features we have

clearly set these parameters.

The levels of the attributes are combined to form concepts, which describe

different potential descriptions of the universal service offer. These are then

presented in pairs to the respondents who are asked which service they

prefer.

The questions used do not represent specific questions we want to answer or

include particular concepts / offers of interest, but are structured to form a

balanced design. The concepts are designed so that not only is each level of

each feature shown a similar number of times, but so that the combinations of

levels across different features is also similarly well balanced. The aim is to

cover as many different combinations as possible, so that the resultant model

can examine any combination of features, not just those tested.

To gather as much detail as possible it is desirable to use many different

questions, covering as many different combinations as we can. However,

limited interviewing time and the need to make the exercise as respondent-

friendly as possible means we cannot ask everyone to respond to a complete

set of possible offers. The approach used is to have a number of different

questionnaire designs that use different sets of product offers. These are

balanced within each set, so everyone sees all the different features but are


4 of 21

also balanced across each version so that overall we have covered all

combinations of features.

A total of 72 ‘packages’ were designed covering different combinations of the

six attributes and levels within them. Residential customers were shown two

packages at a time and asked to choose the one that best met their postal

needs. They also had the option to say that neither was acceptable1. Each

residential customer was shown six different pairs of packages, so that across

all 12 packages, residential customers would have been exposed to each of

the elements three to four times.

It should be noted that the quality of service attribute refers to first class post,

so is only shown when we are looking at a two tier service, as is the price

feature since this also refers to the pricing structure of the two tier approach.

When we present a single tier service, the quality of this service has been

fixed 95% arriving within two days and the price has been fixed at 35p. This

has been done so that the exercise can include a measure of the reactions

towards a single tier two day approach but also concentrate on evaluating the

current two tier structure in more detail. Only the two tier system will therefore

vary in terms of price and quality. The exercise has been structured so that

the two tier service remains the primary focus and this type of service appears

more frequently throughout the exercise.

When choosing between concepts there may be occasions when none of the

options presented to the respondent is very acceptable to them. While they

may have a preference for one over the other, the reality is that they would

not choose either of them

By including a ‘none of these’ option to indicate this within the conjoint data

collection we not only avoid spurious choices being made, when in reality

nothing would be chosen, but also enable to modelling of the point at which

items become sufficiently attractive to be chosen. We can there identify not

1 When choosing between packages there may be occasions when neither of the options is

acceptable to them. While they may have a preference for one over the other, the reality is

that they would not choose either of them. The inclusion of the “none” option avoids spurious

choices being made. It also enables the modelling of the level at which elements become

sufficiently attractive to be chosen. We can therefore identify not just which combinations of

features are more or less attractive but which are attractive enough to be acceptable choices

and those which are not


5 of 21

just which combinations of features are more or less attractive but which are

attractive enough to be acceptable choices and which are not.

The analysis not only creates values for each of the features and levels

included, but also calculates a value for the “none of these” option. This

equates to a threshold which we can include when modelling the

attractiveness of different packages, which tells us whether any given

package is above or below this point at which it becomes an acceptable

option.

An example of one version of the conjoint questions follows.


6 of 21

Version 1

V1Q1 Option A Option B Option C

Service 85% of 1st class post arrives within 1 day

98% of 2nd class post arrives within 3 days

Single service

95% of post arrives within 2 days

Cost of postage first class = 45p

second class = 35p Single service = 35p

Collection and delivery of mail 5 days a week (Monday to Friday) 6 days a week (Monday to Saturday)

Neither option is acceptable, I would have to

use an alternative method

Ability to specify evening / Saturday

delivery of items requiring a signature or too large to fit through letterbox


delivery

Sender can specify an evening or Saturday

delivery for an additional fee of £1.00 per item

Additional services available Insured service (£2.95)


Single insured & next day guaranteed delivery

service (£5.05)




85% of 1st class post arrives within 1 day



second class = 38p

first class = 41p

second class = 32p

Collection and delivery of mail 5 days a week (Saturday and 4 weekdays) 6 days a week (Monday to Saturday)



Ability to specify evening / Saturday delivery of items requiring a signature or

too large to fit through letterbox


delivery



Additional services available Single insured & next day guaranteed delivery

service (£5.05)




7 of 21







second class = 42p

first class = 41p

second class = 32p

Collection and delivery of mail 5 days a week (Saturday and 4 weekdays) 5 days a week (Monday to Friday)








delivery




service (£5.05)


Service Single service

95% of post arrives within 2 days



Cost of postage Single service = 35p first class = 45p

second class = 35p

Collection and delivery of mail 6 days a week (Monday to Saturday) 5 days a week (Monday to Friday) Neither option is acceptable, I would have to



delivery of items requiring a signature or



delivery






service (£5.05)


8 of 21







second class = 38p

first class = 45p

second class = 35p

Collection and delivery of mail 5 days a week (Saturday and 4 weekdays) 6 days a week (Monday to Saturday) Neither option is acceptable, I would have to







delivery

Additional services available Single insured & next day guaranteed delivery

service (£5.05)









second class = 32p

first class = 55p

second class = 42p

Collection and delivery of mail 5 days a week (Monday to Friday) 5 days a week (Saturday and 4 weekdays) Neither option is acceptable, I would have to



delivery of items requiring a signature or





delivery




service (£5.05)


9 of 21

The conjoint analysis

Using attractiveness and acceptability to identify customer needs from a

universal service

The most common use of conjoint is to measure the attractiveness of different

products or services, so that we can model which of them would be chosen.

The total attractiveness of a product is the degree to which people want it –

the more attractive it is, the more they will want it. The one they want the most

is the one they will choose, which in most cases means it is the one they will

buy.

Typically this is because the products/services are on offer in a competitive

environment and the best way to ensure that a product is purchased is to

make sure it is the most attractive option on offer. Customers can choose

whether to buy any product or not to buy at all, so the “none of these” option is

an interesting addition to the modelling process, highlighting situations where

no purchase will be made because nothing is attractive enough to be

acceptable. However, the primary consideration is how attractive a product is

and to degree to which people want to buy it.

In this study we are faced with a slightly different context in which to use the

conjoint results. We can still use the values in exactly the same way as the

any other conjoint based approach, but our objectives are slightly different

from those of many other studies.

Here it is interesting to know how attractive a package is, but this simply tells

us what customers want to be offered and which package they want the most.

Our requirement is to identify what customers truly need to be offered rather

than what they want. This is not just about how attractive an offer is, therefore,

but also about how it relates to the threshold set by the “none of these is

acceptable” option.

For this reason we have chosen to focus on acceptability rather than

attractiveness in evaluating potential service offers. We know that many of the

offers tested are less attractive than the current offer (given that a majority of

the packages that are tested represent changes that reduce the level of

service offered). However, it is possible for a package to be less attractive

than the current offer, but still be attractive enough to be considered


10 of 21

acceptable. In order to focus on what level of service customers need, we

have evaluated service offers in terms of how many customers find them

acceptable or unacceptable (i.e. those who find they do or do not meet their

needs).

What is clear from the data is that the threshold at which the service becomes

unacceptable is, for a majority of respondents, some way below the level of

the current service. This is not to say that customers want a service which is

inferior to the current one, nor that they would be happy to see such a

deterioration in service - they would certainly want to keep as high a level of

service as possible. However, they are prepared to tolerate lower levels of

service, which are still sufficient to be considered an acceptable universal

service, even though they are less attractive that the most desirable offer.

Conjoint output

The choices made in the exercise or preferences for each offer are analysed

to identify what lies behind these choices. The conjoint analysis breaks down

the preferences into the impact that each feature has on that preference.

What we end up with is a set of values (known as utilities) which explain the

choices that customers make.

A utility value is calculated for each level of each feature, and is essentially a

measure of how attractive that level is. Utilities are not measured on a

particular scale, but are relative measures – they indicate how much more or

less attractive one level is compared with another.

From the values of each level, it is also common in conjoint analysis to

calculate the importance of each feature. This is determined by the ability of a

feature to influence customers in their final choices – if there is a bigger

difference in the attractiveness of the levels of a feature, then that feature will

potentially exert a greater influence on the attractiveness of the final service

offer. It will, therefore, be reported as being more important.

Examples of how the conjoint output is reported are shown below. The

importance of features is traditionally presented as a percentage, which may,

therefore, be shown like this, revealing that the choice between the one and

two tier service has had most influence on the overall choices


11 of 21

Importance of features - DUMMY DATA

10.7

12.2

13.5

18.9

20.0

24.8

0.0 5.0 10.0 15.0 20.0 25.0 30.0

Ability to specify evening /Saturday delivery

Additional delivery options

Collection and deliveryservice

Price

Quality of service: 1st classpost

1 v 2 tier service

Within attributes the values of the levels may be viewed like this. It is

important to remember here that the figures do not represent values on

a specific scale – they are relative measures.


12 of 21

Relative attractiveness of service levels / price l evels - DUMMY DATA

0

80

120

110

100

60

00

20

40

60

80

100

120

140

75% 85% 93% 41p / 32p 45p / 35p 49p / 38p 55p / 42p

From the chart above we can see the relative value of different levels of

service. As we would expect there is a drop in attractiveness as quality of

service declines or price increases, but what is important about these values

is that we can see how the different levels compare. We can see that dropping

quality of service from 85% to 75% produces twice as big a decline in

attractiveness than dropping from 93% to 85%. Similarly, one price increase

from 41/32p to 45/35p has little impact, but a second increase to 49/38p has a

much more dramatic effect on perceptions of the service.

We can also compare across attributes, and see that dropping the quality of

service to 85% has a bigger negative impact than increasing the price to

45/35p. We could infer from this, therefore, that it would be better to maintain

the level of service and increase the price, than to keep prices to current

levels and allow quality to decline.

This type of trade-off between the levels of the features is what we need to

know to be able to answer key questions on what service should be offered.

While using the summary numbers to indicate what might happen is

interesting, this is where the conjoint model is of much more use.


13 of 21

Conjoint model

A conjoint model takes the data and uses it to simulate what happens under

different conditions. Rather than simply inferring from the data what the

correct approach would be, we can actually test this and see the impact of

different potential changes on customer perceptions of the service.

The model is an easy to use Excel based tool that allows the user to define a

number of services and identify which is preferred. It also identifies scenarios

where no option which is offered is sufficiently attractive and can also model

those saying that nothing on offer is acceptable.

The utility levels and attribute importance scores provide good summary

measures of what happens in the market, and they offer an excellent way of

contrasting different customer groups but, they are not the best way of making

decisions on the future direction of the service. For this, a better approach is

to construct a model using the conjoint data.

The summary scores generated by the model can hide differences between

respondents – for example we can except that they would all rather have a

service six days a week, but whether or not a Monday to Friday service is

better than retaining Saturday services and dropping a weekday is a matter of

personal opinion. The average scores can disguise this level of difference

between individuals.

In the following screenshot from the demo model, we are testing four possible

strategies – raising prices, dropping the Saturday service, reducing the quality

of service and having a single service. The model shows the percentage that

would choose each option.


14 of 21

Choice (%)


15 of 21

The number of offers included in the model can be varied, so it can also be

used to answer simpler questions. In the following example we test just two

offers - a single tier two day service against the current service offer and see

that it is chosen by a lower percentage.


16 of 21

Choice (%)


17 of 21

We can then change the current service and see how this impacts on the

choice. If, as shown in the following example the choice was between a much

more expensive two tier service or a single tier two day service, then opinions

start to change and the single tier two day service has a slight advantage.


18 of 21

Choice (%)


19 of 21

From this sort of analysis we can see that for many people, the two tier

service is very important and that they are prepared to pay considerably more

to retain it, and yet there are others for whom the idea of a single tier two day

service is quite appealing, even when compared to the current two tier service

and prices.

We can even use the model to test a single tier two day service offer – in the

next example we look at reactions to offering a single tier two day service. We

see that most of those who preferred the two tier approach accept this

(although we know they didn’t want it) but do also see a sizeable increase in

those who view this offer as unacceptable.


20 of 21

Choice (%)


21 of 21

The model allows this type of simulation both in total and by subgroup, so the

appeal and acceptability of service offers can be evaluated for key customer

segments.

calibri font

Documents

delivery service

single service single

class quality of service

choices people

conjoint exercise respondents

delivery of items

class post

quick delivery