Subject: Business
The purpose of this exercise is to:
(a) Provide some skill development in utilising aggregated consumer purchase data using Excel
(a) Provide some skill development in utilising aggregated consumer purchase data using Excel
(b) Better understand the broader concept of competitive structure, the “Duplication of Purchase law” and the idea of market partitions – that is, pockets of the market comprising certain brands that compete more intensely with each other and less intensely with the rest of the market.
(c) Use this understanding to consider a marketing issue. The brand manager for The Radio Network brand is considering a line extension. Into which sub-market should it be launched?
This exercise takes you through the process of analysing a purchase duplication table of real consumer goods data. It uses data on ?�radio listening’ from a New Zealand radio market. What is special about this data set is that it comprises some particular ?�radio listening’ genres – music and talk. Therefore the data provides the opportunity to apply the Duplication of Purchase law to brand-level competition but also to examine the broader competitive level of product type. With the different genres it is more likely there will be some sub-markets or what are called ?�partitions’ that we can identify and confirm using the data.Click Here To Get More On This Paper!!!!
Your task is to work through the exercise and re-create the Excel worksheets shown in the ?�screen shots’ (you don’t have to re-create the written notes on them). Answer the numbered questions that appear through the exercise.
The assignment is due Week Ten (to be advised). The assignment is to be submitted through Turnitin no later than 4-30pm (to be advised). The hard copy of your assignment is to be handed in to your tutor at the start of your tutorial on (to be advised). Please write this assignment in 12 point font, 1.5 spacing. You will need to reference where appropriate. Further information can be found on Webct/Blackboard.
Question 1: What appears to happen to the average levels of duplication as you look across the columns from left to right? Answer this in words to being with.
We can also be more precise and express the answer quantitatively – by using numbers in the answer.
To answer this question quantitatively, we can begin by constructing a graph of penetration versus average duplications. Do this now using the following instructions:Click Here To Get More On This Paper!!!!
1. Request a scatterplot of the data in excel. Ensure that penetration is the x axis and duplication is the y axis. (See screen shot 3 over the page). The scatterplot shows a strong linear relationship between the two variables.
2. Insert a ?�trendline’ (chart -> insert trendline) and request a linear trendline. In ?�options’ ask for the equation and R2 to appear on the chart and ask for the trendline to go through the intercept.
We can also be more precise and express the answer quantitatively – by using numbers in the answer.
To answer this question quantitatively, we can begin by constructing a graph of penetration versus average duplications. Do this now using the following instructions:Click Here To Get More On This Paper!!!!
1. Request a scatterplot of the data in excel. Ensure that penetration is the x axis and duplication is the y axis. (See screen shot 3 over the page). The scatterplot shows a strong linear relationship between the two variables.
2. Insert a ?�trendline’ (chart -> insert trendline) and request a linear trendline. In ?�options’ ask for the equation and R2 to appear on the chart and ask for the trendline to go through the intercept.
Question 2: What is your interpretation of this result? Explain what “duplication = 1.1 x penetration” means.
More analysis
The next step is to more formally construct a model that more fully accounts for the broad pattern exhibited by these data.
To do this, we calculate a statistic called the “purchase duplication coefficient”. The duplication coefficient is a multiplier. It expresses the expected relationship between the size of a brand A and the average proportion of other-brand users that would be expected to also buy brand A in a time period.
We firstly calculate the duplication coefficient for our entire table of purchase duplications. In a sense we have done this already in the preceding step using the graph, which produced a figure of 1.1. But we do it again now using the following formula, which becomes the basis for some more detailed calculations later.
More analysis
The next step is to more formally construct a model that more fully accounts for the broad pattern exhibited by these data.
To do this, we calculate a statistic called the “purchase duplication coefficient”. The duplication coefficient is a multiplier. It expresses the expected relationship between the size of a brand A and the average proportion of other-brand users that would be expected to also buy brand A in a time period.
We firstly calculate the duplication coefficient for our entire table of purchase duplications. In a sense we have done this already in the preceding step using the graph, which produced a figure of 1.1. But we do it again now using the following formula, which becomes the basis for some more detailed calculations later.
Question 3: What is the average duplication and average penetration for the brands in this table?
Question 4: What is the purchase duplication coefficient for this data table?
Why are we doing this?
The reason we are doing this is to create a summary statistic, namely the ?�duplication coefficient’ which is a useful summary of the extent of purchasing sharing, alternatively called switching, in the market. Also, we are doing this to create ?�expected’ purchase duplications for each combination of brands. These ?�expected’ duplications are useful to identify exceptions – brands that are possibly ?�partitioned’, with higher than expected duplications or lower than expected duplications.
We know from our previous graph that the average duplication for a brand is around 1.1 its penetration, except for one brand – which would be somewhat higher.
Using the formula on the previous page, we produce a duplication coefficient of 1.1.
Question 5: This method produces a figure of 1.1. When we used the scatterplot method we also got 1.1. Now, interpet what the figure of 1.1 means.
We can now use this 1.1 figure to create ?�expected’ duplications for each brand. What this means is, we can say ?�given the overall amount of sharing in the market and the size of this brand, we expect X percent of any other brand’s buyers to also buy brand Z’.Click Here To Get More On This Paper!!!!
The way we do this is to multiply the penetration for each brand by the purchase duplication coefficient which gives us the ?�expected’ duplication for each brand. Do this now. Then compute the difference between the actual average duplication and the expected duplication. Refer to screen shot 5 if you need to.
Question 6: Why might we normally expect (in a market with no marked functional differences between the brands) that the duplications down each column are approximately the same?
Given that we are analysing a broad ?�market’ with at least two identifiably different product types, we may well expect some partitioning. This suggests that our first-stage model that simply orders all brands by size is inadequate, because it results in a fair amount of error between our estimated duplications and the actual duplications. Also, one station is not fitting that well to our overall model. Further adjustments need to be made to accommodate these structural features of the market. Therefore we re-order the table according to product type – biggest first, and then in order from the largest to smallest brand within each product type. In other words, have the music stations in order of size, then the talk stations in order of size.Click Here To Get More On This Paper!!!!
Do this now.
Then calculate the average of the duplications for each of the four groups of duplications (e.g. the average of the music -> music duplications, the average of the talk -> talk duplications and so on – these total to 2 x 2 = four combinations). Insert these into an appropriate cell under each group of duplications.
Question 7: Interpret the point raised above. What is the significance of the higher duplication proportions in the music -> music cells compared to the music -> talk cells and the talk -> music cells?
The next way of summarising the data is to now calculate the ?�duplication coefficient’ for each of the product types. We have two product types and we want to know what the relationship is between penetration and duplication for each of these product types, and with each other product type. So we need four duplication coefficients, for example music -> music; talk -> talk; and so on.
Question 8: Interpet the meaning of the figure of 1.4 under the ?�Talk’ column. Model Fit
By calculating duplication coefficients for each product-type pair (music -> music, talk -> talk and so on) we effectively have constructed a ?�model’ of this market and can use it to generate estimated duplications for each brand-pair. Comparing these model-estimated figures to the actual duplications results in a mean absolute deviation of approximately 5 percentage points. Stated more simply, our model estimates the expected duplications to within 5 percentage points of the actual figure, on average. So our slightly more complicated model has enabled us to create quite accurate ?�predicted’ or ?�estimated’ purchase duplications based on the size of the brand, as well as catering to the heightened levels of competition between some of the product types. The original model based simply on ordering by size had an average error of 12 percentage points.
Question 9: Imagine you are the marketing manager for The Radio Network. You are considering launching a new radio station called Radio Hauraki, and you could readily launch it as either a Music, or Talk station. Which options would you choose? To keep the issue simple, assume the margins and
profitability are the same in each sub-market. Thanks to. Professor Malcolm Wright (Massey University) for the original question.Click Here To Get More On This Paper!!!!
In summary we can say:
1. Overall the proportion of buyers of one brand (of any product type – music or talk) who buy another brand (of any type) is mainly a function of the size of that other brand.
2. That said there is distinct partitioning between some of the different product types. Listeners of one genre are much more likely to listen to other stations of the same genre compared to other genres given the size of those other genres.
3. The structure of the market can be put into somewhat more precise numerical terms. For example, the proportion of music listeners who also listen to talk stations is around 0.8 times the penetration levels of those talk stations. This compares to the figure for the proportion of music listeners who will also listen to music stations as being around 1.3 times the penetration level of the music stations.
4. In terms of understanding the extent to which a brand of music or talk station will gain or lose listeners to other stations, a music station will share listeners with other music stations more than would be expected, given their size.
Overall the managerial implications of this analysis are that:
1. This approach helps the manager to understand how a market ?�works’ in terms of which other brands – and which other product types they share customers (or more correctly, customer purchases/listen to) with over time.Click Here To Get More On This Paper!!!!
Methodological note
Note that there is an argument that we cannot always infer the directness of competition from simply examining brand switching (e.g. Lattin and McAlister; Ratneshwar and Shocker, see list over the page). The principal reason for this is that data on switching ignores the impact of occasion – that is, the need that may be being satisfied by the brand on that occasion. For example, suppose a listener listens to a talk station at home and a music station on the way to work. One could argue these two stations are not competing as substitutes for each other, because they are listened to for different reasons. (Note also this
is an argument that is similar to the stated limitations of consumer panel data – namely that because the data represent households, there is the danger of misinterpreting brand switching as an indicator of competition because the brand switch may reflect purchases for different members of the household).
However, in this case I believe that inferring competition purely from switching is reasonably valid. Here we are viewing competition as the extent to which buyers of one brand, or product type also buy other brands/products that (a) share the same broad product characteristics; (b) to which from our own consumption behaviour we know are consumed for approximately the same sorts of reasons; and (c) can be access at the push of a button!
Selected Reference list: Competitive Market Structure. This list is not meant to be exhaustive.
“Repeat Buying – Facts Theory and Applications” by Andrew Ehrenberg. Available as Vol. 5, Journal of Empirical Generalisations in Marketing Science 2000. See www.empgens.com. Among many other contributions, this text outlines the ?�duplication of purchase law’ – that purchase sharing generally falls in-line with brand size unless there are clear functional differences between the brands.
“Understanding Brand Performance Measures” by Ehrenberg, Uncles and Goodhardt. Journal of Business Research Vol. 57, 2004. This paper provides an overview of the fundamental patterns in buyer behaviour arising from the ?�Dirichlet’ model. It further re-inforces the ubiquity of Dirichlet-type patterns, one of which is the duplication of purchase law.Click Here To Get More On This Paper!!!!
“A Probabilistic Model for Testing Hypothesized Hierarchical Market Structures” by Grover and Dillon, Marketing Science Vol. 4, 4, 1985. This paper tests various structures for the ground and instant coffee market (which are analysed separately). It shows how switching in this market is structured according to functional differences in the product (caffeine content, formulation).
“Diversity in analysing brand-switching tables: The car challenge” by Colombo, Ehrenberg and Sabavala, Canadian Journal of Marketing Research Vol. 19, 2000. This paper shows how different forms of analysis may be used for contingency-table brand switching data. It finds support for the proposition that switching is proportional to share, moderated by functional differences between the brands.
“Using a variety-seeking model to identify substitute and complementary relationships among competing products” by Lattin and McAlister, Journal of Marketing Research Vol. 27, August 1985. This paper discusses how inferences about brand competition derived from brand-switching may be confounded by consumer variety seeking. It develops a model to mitigate this problem. The model is, however, very complex.Click Here To Get More On This Paper!!!!
“Competitive Market Structures: A Subset Selection Approach” by Kannan and Sanchez, Management Science Vol 40, 11, 1994. This paper analyses two consumer good markets, namely coffee and flavoured crackers. It develops an approach for analysing competitive structures and testing for variety seeking. It also features a very clear pictorial approach for presenting the results for a partitioned market.
“Substitution in Use and the Role of Usage Context in Product Category Structures” by Ratneshwar and Shocker, Journal of Marketing Research Vol. 28, August 2001. The paper examines how the common vs distinctive features of products interact with usage situations to affect substitutability in use. Among other findings, the paper concludes that common attributes have a stronger association with similarity than do common usages (p. 286). For (my own) example: a person might use chocolate or popcorn to eat in a movie (similar usage) but there is little commonality in the attributes of the two products, so they are not seen as similar.
“Does the Duplication of Viewing Law apply to Radio Listening?” by Gavin Lees and Malcolm Wright European Journal of Marketing. Vol 47 Iss 3 (Date Online 28/5/2012). This paper examines the long standing interest in the duplication of audience between media vehicles and finds that the Duplication of Listening does broadly follow the Duplication of Viewing Law. Contrary to popular belief most of the deviations from a mass market are not due to micro-formats (e.g. classic rock) but rather are explained by a broad partitioning of the market between ?�talk’ and ?�music’ segments, although they do identify a unique station that still deviates from its parent partition.
Why are we doing this?
The reason we are doing this is to create a summary statistic, namely the ?�duplication coefficient’ which is a useful summary of the extent of purchasing sharing, alternatively called switching, in the market. Also, we are doing this to create ?�expected’ purchase duplications for each combination of brands. These ?�expected’ duplications are useful to identify exceptions – brands that are possibly ?�partitioned’, with higher than expected duplications or lower than expected duplications.
We know from our previous graph that the average duplication for a brand is around 1.1 its penetration, except for one brand – which would be somewhat higher.
Using the formula on the previous page, we produce a duplication coefficient of 1.1.
Question 5: This method produces a figure of 1.1. When we used the scatterplot method we also got 1.1. Now, interpet what the figure of 1.1 means.
We can now use this 1.1 figure to create ?�expected’ duplications for each brand. What this means is, we can say ?�given the overall amount of sharing in the market and the size of this brand, we expect X percent of any other brand’s buyers to also buy brand Z’.Click Here To Get More On This Paper!!!!
The way we do this is to multiply the penetration for each brand by the purchase duplication coefficient which gives us the ?�expected’ duplication for each brand. Do this now. Then compute the difference between the actual average duplication and the expected duplication. Refer to screen shot 5 if you need to.
Question 6: Why might we normally expect (in a market with no marked functional differences between the brands) that the duplications down each column are approximately the same?
Given that we are analysing a broad ?�market’ with at least two identifiably different product types, we may well expect some partitioning. This suggests that our first-stage model that simply orders all brands by size is inadequate, because it results in a fair amount of error between our estimated duplications and the actual duplications. Also, one station is not fitting that well to our overall model. Further adjustments need to be made to accommodate these structural features of the market. Therefore we re-order the table according to product type – biggest first, and then in order from the largest to smallest brand within each product type. In other words, have the music stations in order of size, then the talk stations in order of size.Click Here To Get More On This Paper!!!!
Do this now.
Then calculate the average of the duplications for each of the four groups of duplications (e.g. the average of the music -> music duplications, the average of the talk -> talk duplications and so on – these total to 2 x 2 = four combinations). Insert these into an appropriate cell under each group of duplications.
Question 7: Interpret the point raised above. What is the significance of the higher duplication proportions in the music -> music cells compared to the music -> talk cells and the talk -> music cells?
The next way of summarising the data is to now calculate the ?�duplication coefficient’ for each of the product types. We have two product types and we want to know what the relationship is between penetration and duplication for each of these product types, and with each other product type. So we need four duplication coefficients, for example music -> music; talk -> talk; and so on.
Question 8: Interpet the meaning of the figure of 1.4 under the ?�Talk’ column. Model Fit
By calculating duplication coefficients for each product-type pair (music -> music, talk -> talk and so on) we effectively have constructed a ?�model’ of this market and can use it to generate estimated duplications for each brand-pair. Comparing these model-estimated figures to the actual duplications results in a mean absolute deviation of approximately 5 percentage points. Stated more simply, our model estimates the expected duplications to within 5 percentage points of the actual figure, on average. So our slightly more complicated model has enabled us to create quite accurate ?�predicted’ or ?�estimated’ purchase duplications based on the size of the brand, as well as catering to the heightened levels of competition between some of the product types. The original model based simply on ordering by size had an average error of 12 percentage points.
Question 9: Imagine you are the marketing manager for The Radio Network. You are considering launching a new radio station called Radio Hauraki, and you could readily launch it as either a Music, or Talk station. Which options would you choose? To keep the issue simple, assume the margins and
profitability are the same in each sub-market. Thanks to. Professor Malcolm Wright (Massey University) for the original question.Click Here To Get More On This Paper!!!!
In summary we can say:
1. Overall the proportion of buyers of one brand (of any product type – music or talk) who buy another brand (of any type) is mainly a function of the size of that other brand.
2. That said there is distinct partitioning between some of the different product types. Listeners of one genre are much more likely to listen to other stations of the same genre compared to other genres given the size of those other genres.
3. The structure of the market can be put into somewhat more precise numerical terms. For example, the proportion of music listeners who also listen to talk stations is around 0.8 times the penetration levels of those talk stations. This compares to the figure for the proportion of music listeners who will also listen to music stations as being around 1.3 times the penetration level of the music stations.
4. In terms of understanding the extent to which a brand of music or talk station will gain or lose listeners to other stations, a music station will share listeners with other music stations more than would be expected, given their size.
Overall the managerial implications of this analysis are that:
1. This approach helps the manager to understand how a market ?�works’ in terms of which other brands – and which other product types they share customers (or more correctly, customer purchases/listen to) with over time.Click Here To Get More On This Paper!!!!
Methodological note
Note that there is an argument that we cannot always infer the directness of competition from simply examining brand switching (e.g. Lattin and McAlister; Ratneshwar and Shocker, see list over the page). The principal reason for this is that data on switching ignores the impact of occasion – that is, the need that may be being satisfied by the brand on that occasion. For example, suppose a listener listens to a talk station at home and a music station on the way to work. One could argue these two stations are not competing as substitutes for each other, because they are listened to for different reasons. (Note also this
is an argument that is similar to the stated limitations of consumer panel data – namely that because the data represent households, there is the danger of misinterpreting brand switching as an indicator of competition because the brand switch may reflect purchases for different members of the household).
However, in this case I believe that inferring competition purely from switching is reasonably valid. Here we are viewing competition as the extent to which buyers of one brand, or product type also buy other brands/products that (a) share the same broad product characteristics; (b) to which from our own consumption behaviour we know are consumed for approximately the same sorts of reasons; and (c) can be access at the push of a button!
Selected Reference list: Competitive Market Structure. This list is not meant to be exhaustive.
“Repeat Buying – Facts Theory and Applications” by Andrew Ehrenberg. Available as Vol. 5, Journal of Empirical Generalisations in Marketing Science 2000. See www.empgens.com. Among many other contributions, this text outlines the ?�duplication of purchase law’ – that purchase sharing generally falls in-line with brand size unless there are clear functional differences between the brands.
“Understanding Brand Performance Measures” by Ehrenberg, Uncles and Goodhardt. Journal of Business Research Vol. 57, 2004. This paper provides an overview of the fundamental patterns in buyer behaviour arising from the ?�Dirichlet’ model. It further re-inforces the ubiquity of Dirichlet-type patterns, one of which is the duplication of purchase law.Click Here To Get More On This Paper!!!!
“A Probabilistic Model for Testing Hypothesized Hierarchical Market Structures” by Grover and Dillon, Marketing Science Vol. 4, 4, 1985. This paper tests various structures for the ground and instant coffee market (which are analysed separately). It shows how switching in this market is structured according to functional differences in the product (caffeine content, formulation).
“Diversity in analysing brand-switching tables: The car challenge” by Colombo, Ehrenberg and Sabavala, Canadian Journal of Marketing Research Vol. 19, 2000. This paper shows how different forms of analysis may be used for contingency-table brand switching data. It finds support for the proposition that switching is proportional to share, moderated by functional differences between the brands.
“Using a variety-seeking model to identify substitute and complementary relationships among competing products” by Lattin and McAlister, Journal of Marketing Research Vol. 27, August 1985. This paper discusses how inferences about brand competition derived from brand-switching may be confounded by consumer variety seeking. It develops a model to mitigate this problem. The model is, however, very complex.Click Here To Get More On This Paper!!!!
“Competitive Market Structures: A Subset Selection Approach” by Kannan and Sanchez, Management Science Vol 40, 11, 1994. This paper analyses two consumer good markets, namely coffee and flavoured crackers. It develops an approach for analysing competitive structures and testing for variety seeking. It also features a very clear pictorial approach for presenting the results for a partitioned market.
“Substitution in Use and the Role of Usage Context in Product Category Structures” by Ratneshwar and Shocker, Journal of Marketing Research Vol. 28, August 2001. The paper examines how the common vs distinctive features of products interact with usage situations to affect substitutability in use. Among other findings, the paper concludes that common attributes have a stronger association with similarity than do common usages (p. 286). For (my own) example: a person might use chocolate or popcorn to eat in a movie (similar usage) but there is little commonality in the attributes of the two products, so they are not seen as similar.
“Does the Duplication of Viewing Law apply to Radio Listening?” by Gavin Lees and Malcolm Wright European Journal of Marketing. Vol 47 Iss 3 (Date Online 28/5/2012). This paper examines the long standing interest in the duplication of audience between media vehicles and finds that the Duplication of Listening does broadly follow the Duplication of Viewing Law. Contrary to popular belief most of the deviations from a mass market are not due to micro-formats (e.g. classic rock) but rather are explained by a broad partitioning of the market between ?�talk’ and ?�music’ segments, although they do identify a unique station that still deviates from its parent partition.
No comments:
Post a Comment