Time to add a new skill to your market research tool kit or grab a quick refresher? Tim Bock will show you everything you need to know in this live 20 min webinar. 💪

What you'll learn in this 20 min video:

- When to use TURF, factor, and correspondence analysis
- How to set up the data and interpret the analysis
- How to share the results with more exciting, inspiring, & visual reports.

I'm going to walk you through techniques that are useful for designing product portfolios: correspondence analysis, factor analysis, and, TURF.

I'm going to be showing you everything from Displayr, but you can get all the same outputs in Q, other than the interactive venn diagram.

**Case study: Bubble Gum Flavor Portfolio Optimization**

A bubble gum company has two flavors: classic, which is what they call Grape, and cola. It's decided to expand its portfolio. Which flavors should it add?

This same problem exists in lots of other area:

- Which range of products should a retailer sell?
- Which media should we use?
- Which product features should be offered?

**Raw Data**

I'll start by adding some data

*Add data set > TURF > Flavors.sav*

We have data from 712 people.

We asked people which of 11 flavours they would like.

*Table > Raw Data > Variable Set*

For each person, a 1 indicates if they like the flavor and a 0 indicates that they don't.

So, for the first person, we can see they like Classic Bubble Gum, Group, Sour, Orange, and Cola.

**Reach by Flavor**

7. 66% of people like the Classic Bubble Gum. To use a bit of jargon, this is the Reach of Classic. That is, if we just offer Classic, our portfolio reaches 66% of the market.

Cola, our client's second brands, has a reach of 38%

If you add these numbers together, 38 + 66 = 104%, which is impossible. Naively adding up the numbers ignores that there is some duplication, which is the jargon for people that like both flavors. We need to compute the Total Unduplicated Reach of the flavors.

*Combine > Create as New Category*

Classic + Cola have a total unduplicated reach of 75%.

Or, 74.7% to be more precise.

Our goal in this study is to find the best combinations of flavors to grow reach.

*Duplicate page > Title: Duplication Matrix*

The most basic way of analyzing data like this is to look at a duplication matrix, which is jargon for a table that contains the brands in both the rows and columns.

By default, Display shows column percentages, but I prefer to look at total percentages with an analysis like this.

*Statistics - Cells*

*Total %*

*Turn off Column %*

Reading across the first row, we can see that it says that 66% of people like Classic, which is what we saw before.

35% like Super-Strong both and Classic.

45% like Grape and Classic, and so on.

Remember before we saw that the reach was 75%, for Cola and Classic? We can compute that using the numbers on this table.

Classic has a 65.9% reach.

Cola has 37.5%

But the duplication between the two is 28.7%. That is, 29% of the sample like both.

So the unduplicated reach is 75%%

There's a lot of numbers on this table and if we had a lot of time and an amazing brain, we could find the patterns.

But, if like me you're in a rush, we need a shortcut. And, the standard shortcut for quickly findings patterns is correspondence analysis

**Correspondence analysis of a Duplication Matrix**

The way to find out where a technique is in Displayr is to type what you want into the search box.

*Search: Correspondence analysis*

Displayr uses this yellow shading to guide you.

*... > Show option*

As you can see, there are three options? Which should we choose?

We've built some smarts into the search.

So, rather than type the technique, I'm going to tell it I've got a duplication matrix, and see what it suggests

*Search: Duplication matrix*

Ah, it's recommending correspondence analsysi of a square table, where square in this case means with the same labels in both the rows and the columns

*Select Correspondence Analysis of a Square Matrix*

*Input table: table.flavours.by.flavours*

The closer two flavors are together, the higher their duplication.

Our current flavours are Cola and Classic Bubble Gum.

So, there's a high level of duplication between our two current flavors of Cola and Classic, so there's probably no point in adding super strong or grape, as they also appeal to the same core group of users

All else being equal, we should offer Watermelon, as it's the furthest from our current brands, so will have the least duplication.

But, all else isn't equal.

It only makes sense to add Watermelon if it has low duplication and if it is popular. So, let's show the popularity of each flavor. That is, it's reach, as a circle.

*Output: Bubble chart*

*Bubble sizes: Table.flavors.2*

Ah, that's unfortunate. Watermelon's got a small circle, so it's not a popular flavor.

But, lastly, note that the biggest circles are all for the original flavor segment, so it may make sense to have multiple flavors from here.

Correspondence analysis compresses data into two dimensions. We can get a more nuanced read by using factor analysis.

**Factor Analysis**

As I'm sure most of you know, what we typically call factor analysis in market research, is technically principal component analysis

*Insert < Dimension Reduction > Principal Components Analysis and drag across Flavors.*

This reveals three dimensions of flavor

- The first shows sweeter fruits
- The second is the traditional bubble gum flavors
- The third is citrus fruits

This suggests that a good portfolio should represent each of these dimensions.

Display's factor analysis automatically works out the number of components based on the kaiser rule.

It's usually worthwhile to have a look to see if this is clearly the best solution, or, there's a bit of noise. In factor analysis, this is done using a scree plot.

The x axis shows the number of components or dimensions.

The y axis shows how much information they explain. If there was a sharp corner, it would tell us that the three factor solution was clearly the best. But, it's pretty smooth here, telling us we can explore other solutions.

*Output: Loadings table.*

Rule for selecting components: Number of components

With two component, we basically get the same result as with the correspondence analysis

We find a dimension of fruit and another showing classics.

And sour and chocolate are in-between.

Let's explore the four factor solution

Number of components: 4

OK, so this is very similar to the three factor solution

We have fruits

We have the classic flavors

We have citrus

And, chocolate's been pushed out on its own.

**Total Unduplicated Reach and Frequency: 2 Alternatives (Including Cola and Classic)**

Correspondence analysis and factor analysis are general techniques that help to find patterns in data. There's a special tool for portfolio planning, called TURF.

*Anything > Advanced >TURF > TURF Analysis*

The way it works is it searches through the data and finds the combinations of alternatives that maximize reach.

This table is showing us the 10 best portfolios that consist of two flavours.

It's found that if we were to have two flavours, the best would be classic and strawberry, and they would reach 82.4% of the market, with a total of 747 choices, where this frequency counts people twice if they like each of the flavours.

With three alternatives, we are seeing that we get one from the classic segment, one from the citrus, and one from sweet fruits, which is basically what we saw before.

But note that the second best portfolio, which is almost as good, has Classic and Grape, where we saw before have high duplication.

These portfolios do not differ a lot, with the best portfolio being only .6% above the second best. This is good news, as it means that we have a fair bit of flexibility in choosing our best portfolio, and can trade off tactical and operational factors.

If we are going to have more than two flavors, it's likely that we would want to include our current classic and cola flavors.

We can run the TURF with these and other types of constraints.

*CONSTRAINTS > Must include: Classic Bubble Gum*

*Must include: Cola*

Let's set this back to a portfolio size of 2 to set the baseline.

We again see the reach at 75%

*Number of alternatives: 3*

So, if we were to add one flavour, it would be strawberry, which is consistent with what we saw before.

*Number of alternatives: 4*

With four flavors we get Sour added in.

*Number of alternatives: 5*

Now Grape is added to the mix

*Number of alternatives: 6*

With 6 flavours, Chocolate final enters. But, we've lost sour which has been replaced with Orange. If you look down you can see that the 3rd best portfolio includes Sour, and the differences is very marginal at 0.3%. This is well within noise, and we can force the results to be consistent by making Sour a constraint:

Let's put it together

**Incremental Gains of New Flavors**

An incrementality plot or waterfall is a great way of showing the effect of adding to portfolio size.

*Anything > Advanced Analysis > TURF > Incrementality Plot*

So, adding Strawberry gains us an extras 14% of reach.

Then adding Sour another 6%.

If we then add in Grape, that gives an extra 2.5%. But given, that Classic is a grape flavor, there's not much of a gain here, so the trick seems to be to add just strawberry and Sour and round out the portfolio at 4 flavors.

Read more