Product Availability (Multi-store)

 

Differential product availability in the marketplace is one of the most common reasons that simulated shares of preference don't match actual market shares.  Perhaps Brand A is only available in half the stores (or regions) of a country, but Brand B is available in 95% of the stores (or regions).  Fortunately, if you have appropriate data regarding which products are available in which stores (or regions), then there is a straightforward and proper way called Multi-Store simulations to adjust simulated shares of preference to account for this reality.  

 

When you go to the My Scenario Settings tab and check Apply Product Availability (Multi-store), you can click the icon to edit the settings for this option.

 

Simulated shopping trip count per respondent

 

This option controls how many "shopping trips" each respondent takes.  In each shopping trip, the respondent is assumed to visit a specific store (or region) and to purchase only among the products available within that store.  In the next shopping trip, the respondent might be assigned to visit a different store.  To stabilize the simulated share of preference results, we recommend using at least 500 trips per respondent.

 

Note: if using the default Randomized First Choice simulation method, each respondent is simulated multiple times via Randomized First Choice iterations (where each iteration involves a new draw of attribute level error).  By default, we perform 3 shopping trips (if applying Multi-store adjustment for Product Availability) per Randomized First Choice iteration.  So, if a respondent gets 500 Randomized First Choice iterations, that respondent would make 500 x 3 = 1,500 total shopping trips.

 

Apply average product availability in the market

 

If you only know on average in what percent of the stores (regions) each product is available, then this is the only Multi-store option available to you.  However, if you have more information about specifically which products are available in which stores (regions), it is better to use the granular store option described further below.

 

How the average product availability option works: imagine there are four products available within a market scenario, products A, B, C and D.  Further imagine that the only thing you know is the percent of total stores in which each product is available:

 

A  50%

B  90%

C  95%

D  70%

 

(Note that the probabilities are typed as "50" to indicate 50% rather than 0.50; the software automatically adds the "%" symbol.  You may specify decimal places of precision, such as by typing "50.523" to indicate "50.523%".)

 

Each respondent is sent shopping typically 100s of times.  In each shopping trip, the different products are made available with likelihood equal to the probabilities expressed above.  Over the thousands of shopping trips, respondents will see Product A available 50% of the time, Product B available 90%, etc.  (We ensure that in each shopping trip, at least one of the products is available).  

 

For example, in the first shopping trip, respondent #1 may find that only products B and C are available to purchase.  The share of preference across products B and C (and potentially the None option, if you include this in your study) sums to 100% within that first shopping trip.  Shares of preference for A and D are 0% in that first shopping trip.

 

Apply granular store information

 

If you know which products are available in which specific stores (or regions), then you will obtain better results if you apply this granular store information. Specifically, the patterns of availability of the products across multiple stores inform the simulator regarding the competitive effects among the products due to joint availability (e.g. two products that tended to be correlated in terms of the availability in stores would tend to compete more closely).  

 

Let's imagine you knew that there were eight different stores (regions) and that the availability of products A-D (that you have specified for the current market scenario) within those stores was as follows:

 

 

The Aggregate Store Visit Probability indicates the likelihood that each respondent will visit each of the stores and that column must sum to 100%.  In each store shopping trip (each respondent typically is specified to take 100s of trips), the respondent is randomly assigned to visit one of the eight stores with likelihood specified in the Aggregate Store Visit Probability.  Thus, it is expected that 18% of the trips for each respondent will be to Store 1, 12% to Store 2, etc.

 

When a respondent visits Store 1, all four products are available for purchase.  However, if the respondent visits Store 2, Products A and B are not available and will receive 0% share of preference.  For Store 2 visits, the share of preference (100%) will be apportioned among Brand C, Brand D (and potentially the None option, if it is specified) according to the simulation rule that has been selected (First Choice, Randomized First Choice, Share of Preference).

 

Note that Products A and B are nearly perfectly correlated in their availability across the stores: either both available or both not available in seven of the eight stores,  representing 98% of the purchase volume (Store 7, the only store that breaks from this pattern, accounts for only 2% of sales volume).  This pattern will induce further heightened substitutability between Products A and B, beyond that which may be present in the patterns of respondent preferences.

 

Of course, if we knew which store each respondent was most likely to visit, we wouldn't need to randomly assign respondents to stores across multiple simulated shopping trips; we could assign a respondent specifically to the store(s) that was most probable for that individual.

 

Use per-respondent store visit probabilities

 

If you know which store (region) each respondent is likely to visit, you should apply respondent-specific store visit probabilities, rather than allow respondents to visit all potential stores according to a non-zero probability assignment.  To add per-respondent store visit probabilities, you must prepare a .CSV file with the information (you may use Excel to prepare this file).  For our example above with eight stores in the multi-store simulation, the file has the following format (where we have only shown the data for the first three respondents and first three stores, but the other stores would be found in five additional columns to the right):

 

 

Each respondent is given a probability of visit to each of the stores (regions), which probabilities must sum to 100 across the stores.  To indicate 48.26%, specify 48.26 in the file.  A header row of labels is permitted, but not required.

 

In the example above respondent #1001 has a 48.26% probability of visiting Store 1, a 22.13% probability of visiting Store 2 (and the probabilities for the other six stores account for the remainder, such that the sum across all eight stores is 100%).  Respondent #1002 only can visit Store 2 (Store 2 has a visit probability of 100%).  Following the example from the previous section, respondent #1002 visiting Store 2 only encounters products C & D available for purchase.

 

If you use per-respondent store visit probabilities, the per-respondent probabilities are used for assigning the respondent to visit the different stores rather than the Aggregate Store Visit Probabilities specified in the Product Availability by Stores table.  However, if a respondent is missing from the per-respondent store visit probabilities .CSV file (but has part-worth utilities), the simulator will assign that respondent to visit stores with probabilities proportional to the Aggregate Store Visit Probabilities specified in the Product Availability by Stores table of this dialog.  Thus, the software asks you to specify Aggregate Store Visit Probabilities even though it seems that this information generally wouldn't be used.

 

---

Applying Different Product Definitions (e.g. prices) across Different Stores

 

If different stores set different prices for the same products, at first glance the software doesn't seem to be able to support this; but there is an advanced trick to accomplish it:

 

1.Create multiple products in the scenario that correspond to the unique price specifications available in the market.  For example, if you know that some stores sell your product at $10 and some sell it at $12, you would enter your product twice into the market simulation scenario: once at $10 and once at $12.

 

2.Within the granular store information table, check which products are being offered in which stores.  For the stores offering your product at $10, only specify that the $10 version of your product is available.  For the stores offering your product at $12, only specify that the $12 product is available.

 

3.On the My Scenario Settings tab, Miscellaneous ribbon group, click the icon.  Net (sum and collapse) the two versions of your product (at $10 and at $12) into a single product for reporting purposes.