Abstract
Here a brief recapitulation of the study of brand choice processes is given, and the major conclusions are reported.In chapter 2 we discussed the empirical brand choice data used throughout the study. We saw that these were purchase histories of members of the Dutch Attwood Consumer Panel for the products fopro (a pseudonym), beer and margarine during the years 1967 and 1968. A brief description of the type of information, available for each purchase, was given, and some important characteristics of the markets for each of the 3 products were also reported. In addition some attention was paid to the special brands of the 3 products which are repeatedly distinguished in the study and the source of information for the advertising figures used in the analysis was mentioned.In chapter 3, 5 different brand choice models with the corresponding testing and estimation procedures were presented, viz., the:
1. Homogeneous Bernoulli Model (HOBM);
2. Homogeneous Markov Model (HOMM);
3. Heterogeneous Bernoulli Model (HEBM);
4. Heterogeneous Markov Model (HEMM);
5. Linear Learning Model (LLM).Of these models (1) and (3) assume no influence of former purchases on current brand choice, i.e. no purchase feedback. In these two models it is assumed that a consumer purchases a certain brand with constant probability, which probability - according to (1) - is equal for all consumers, but - in the case of (3) - may vary over consumers.All the other 3 models assume purchase feedback. In the case of the Markov models (2) and (4) the influence of former purchases is limited to the recent purchase occasions and expressed in so-called transition probabilities. For the HOMM these transition probabilities are equal for all consumers; in the case of the (HEMM); different consumers can have different transition probabilities. According to the Linear Learning Model a consumer is assumed to have - at a given purchase occasion - a probability p of buying a certain brand. After a purchase this probability is transformed in a way dependent on the brand bought at that purchase. This implies that in the LLM quite a number of former purchases have an influence, but the influence of the brand choice at an earlier purchase occasion diminishes with the number of purchases made since that occasion.In chapter 4 it was examined how far the brand choice models just mentioned gave a good description of the empirical brand choice processes of fopro, beer and margarine. This was done by carrying out testing procedures and performing a simulation study.From the test results it became clear that in all cases a definite purchase feedback is present; the Bernoulli models HOBM and HEBM did not give a good fit, and the Markov and Learning models also showed an evident influence of previous purchases. Of the homogeneous Markov models the first order Markov model, which is the brand choice model most discussed in the literature, did not give a good fit. The performance of the second order model was better, but was still not really satisfactory. The Heterogeneous Markov Model (HEMM) also did not give a good description of the brand choice processes observed. The brand choice model which gave the best results, was the Linear Learning Model (LLM). In every case this model was superior to all other models.In the simulation study, where the ability of the various brand choice models to reproduce the original brand choice processes was examined, the superiority of the Linear Learning Model again appeared, while the homogeneous Markov models (first and second order) offered a much worse reproduction. A curious point is that the HEBM appeared to give a reproduction of the original brand choice processes which was almost as good as that of the LLM. At first glance this seemed contradictory, but a closer examination of the LLM-parameters showed that these parameters were such that the LLM-processes concerned exhibited a lot of seeming zero-order behavior. Because the HEBM is a zero-order model, this explains the phenomenon observed. Taking the results of the testing procedures and the simulation study together, it appears, that - of all brand choice models used - the Linear Learning Model evidently gave the best description of the empirical brand choice processes.In chapter 5 we briefly discussed a number of learning models from the viewpoint of their application possibilities to brand choice processes. The non-linear operator models treated appeared to offer no great perspectives. The stimulus sampling models, an example of which was applied to the fopro, beer and margarine data looked more promising. Some further properties of the Linear Learning Model were given, in particular relating to equilibrium behavior, which are useful because they can be used to compute long term market shares. It was also shown that the Linear Learning Model can be generalized, so that brand choice processes can be handled, for which not all assumptions of the ordinary LLM hold.In chapter 6 we analysed the empirical brand choice processes with the aid of the variable 'poolsize' introduced there. Poolsize is defined as the 'number of different brands bought during the last 10 purchases'. An important finding is that - in their purchase histories - consumers show periods of routinized buying alternated with periods of brand switching. This is in agreement with the good fit of the LLM observed in chapter 4, because the parameters of the LLM estimated there are such that in the corresponding brand choice processes there will be long periods during which brand switches are very unlikely, alternated with periods in which the probability of moving to another brand is considerable.Further conclusions resulting from the poolsize approach, are that a consumer simultaneously considers a limited number of brands as potential choice candidates and that consumers do not often straightforwardly switch from one brand to another, but usually exhibit search behavior, which accompanies a transition to another brand.In chapter 7 we examined the relationships between brand choice and a number of environmental variables.With respect to shop choice, it was found that brand choice and shop choice are rather closely related. This interdependence cannot completely be traced back to the fact that the choice of a shop simply limits the set of different brands from which a choice can be made; it seems that an autonomous general proneness-to-change factor exists, which means that some consumers show great variation with respect to shops as well as to brands. Moreover, there can be distinguished specific proneness-to-brand- change and proneness-to-shop-change factors. With respect to the effect of the marketing variables the following can be remarked: by means of a multiple regression analysis it was found that for a number of brands there was a significant influence of price and/or advertising (the latter measured by expenditures made) on market share, repeat purchase probability and on the probability of making a transition from another brand to the brand concerned. Further it was found that deal purchases are relatively often associated with brand switches, so that dealing seems to be an instrument having the ability to induce brand changes.As for the effect of inter-purchase times on brand choice, it was found for fopro that the probability of purchasing the same brand as the previous one (= repeat purchase probability) decreases as inter-purchase times become longer. For beer and margarine no effect of inter-purchase times could be established.In chapter 8 we studied the relationship between brand choice behavior and household variables.It was found that socio-economic variables have only weak relationships with brand choice variables. For all 3 products we observed some influence of size of town, region, children and attitude scores in relation to buying behavior. In incidental cases there was also an influence of family size, age of housewife and the possession of a refrigerator and a television set.Between brand choice variables and other purchase variables the relationships are stronger. It was observed that households which, relatively, show a lot of brand switching pay a lower price, make more deal- purchases, make more purchases in self-service shops or supermarkets, buy more in shops belonging to chains, have more variation in interpurchase times and in volume per purchase occasion and buy more different package sizes.To a certain extent brand choice behavior was found to be transitive over products, i.e. to some extent households showed the same type of brand choice behavior in relation to different products, but not to such a degree that a general brand choice behavior could be spoken of which could serve as an independent basis for market segmentation.
1. Homogeneous Bernoulli Model (HOBM);
2. Homogeneous Markov Model (HOMM);
3. Heterogeneous Bernoulli Model (HEBM);
4. Heterogeneous Markov Model (HEMM);
5. Linear Learning Model (LLM).Of these models (1) and (3) assume no influence of former purchases on current brand choice, i.e. no purchase feedback. In these two models it is assumed that a consumer purchases a certain brand with constant probability, which probability - according to (1) - is equal for all consumers, but - in the case of (3) - may vary over consumers.All the other 3 models assume purchase feedback. In the case of the Markov models (2) and (4) the influence of former purchases is limited to the recent purchase occasions and expressed in so-called transition probabilities. For the HOMM these transition probabilities are equal for all consumers; in the case of the (HEMM); different consumers can have different transition probabilities. According to the Linear Learning Model a consumer is assumed to have - at a given purchase occasion - a probability p of buying a certain brand. After a purchase this probability is transformed in a way dependent on the brand bought at that purchase. This implies that in the LLM quite a number of former purchases have an influence, but the influence of the brand choice at an earlier purchase occasion diminishes with the number of purchases made since that occasion.In chapter 4 it was examined how far the brand choice models just mentioned gave a good description of the empirical brand choice processes of fopro, beer and margarine. This was done by carrying out testing procedures and performing a simulation study.From the test results it became clear that in all cases a definite purchase feedback is present; the Bernoulli models HOBM and HEBM did not give a good fit, and the Markov and Learning models also showed an evident influence of previous purchases. Of the homogeneous Markov models the first order Markov model, which is the brand choice model most discussed in the literature, did not give a good fit. The performance of the second order model was better, but was still not really satisfactory. The Heterogeneous Markov Model (HEMM) also did not give a good description of the brand choice processes observed. The brand choice model which gave the best results, was the Linear Learning Model (LLM). In every case this model was superior to all other models.In the simulation study, where the ability of the various brand choice models to reproduce the original brand choice processes was examined, the superiority of the Linear Learning Model again appeared, while the homogeneous Markov models (first and second order) offered a much worse reproduction. A curious point is that the HEBM appeared to give a reproduction of the original brand choice processes which was almost as good as that of the LLM. At first glance this seemed contradictory, but a closer examination of the LLM-parameters showed that these parameters were such that the LLM-processes concerned exhibited a lot of seeming zero-order behavior. Because the HEBM is a zero-order model, this explains the phenomenon observed. Taking the results of the testing procedures and the simulation study together, it appears, that - of all brand choice models used - the Linear Learning Model evidently gave the best description of the empirical brand choice processes.In chapter 5 we briefly discussed a number of learning models from the viewpoint of their application possibilities to brand choice processes. The non-linear operator models treated appeared to offer no great perspectives. The stimulus sampling models, an example of which was applied to the fopro, beer and margarine data looked more promising. Some further properties of the Linear Learning Model were given, in particular relating to equilibrium behavior, which are useful because they can be used to compute long term market shares. It was also shown that the Linear Learning Model can be generalized, so that brand choice processes can be handled, for which not all assumptions of the ordinary LLM hold.In chapter 6 we analysed the empirical brand choice processes with the aid of the variable 'poolsize' introduced there. Poolsize is defined as the 'number of different brands bought during the last 10 purchases'. An important finding is that - in their purchase histories - consumers show periods of routinized buying alternated with periods of brand switching. This is in agreement with the good fit of the LLM observed in chapter 4, because the parameters of the LLM estimated there are such that in the corresponding brand choice processes there will be long periods during which brand switches are very unlikely, alternated with periods in which the probability of moving to another brand is considerable.Further conclusions resulting from the poolsize approach, are that a consumer simultaneously considers a limited number of brands as potential choice candidates and that consumers do not often straightforwardly switch from one brand to another, but usually exhibit search behavior, which accompanies a transition to another brand.In chapter 7 we examined the relationships between brand choice and a number of environmental variables.With respect to shop choice, it was found that brand choice and shop choice are rather closely related. This interdependence cannot completely be traced back to the fact that the choice of a shop simply limits the set of different brands from which a choice can be made; it seems that an autonomous general proneness-to-change factor exists, which means that some consumers show great variation with respect to shops as well as to brands. Moreover, there can be distinguished specific proneness-to-brand- change and proneness-to-shop-change factors. With respect to the effect of the marketing variables the following can be remarked: by means of a multiple regression analysis it was found that for a number of brands there was a significant influence of price and/or advertising (the latter measured by expenditures made) on market share, repeat purchase probability and on the probability of making a transition from another brand to the brand concerned. Further it was found that deal purchases are relatively often associated with brand switches, so that dealing seems to be an instrument having the ability to induce brand changes.As for the effect of inter-purchase times on brand choice, it was found for fopro that the probability of purchasing the same brand as the previous one (= repeat purchase probability) decreases as inter-purchase times become longer. For beer and margarine no effect of inter-purchase times could be established.In chapter 8 we studied the relationship between brand choice behavior and household variables.It was found that socio-economic variables have only weak relationships with brand choice variables. For all 3 products we observed some influence of size of town, region, children and attitude scores in relation to buying behavior. In incidental cases there was also an influence of family size, age of housewife and the possession of a refrigerator and a television set.Between brand choice variables and other purchase variables the relationships are stronger. It was observed that households which, relatively, show a lot of brand switching pay a lower price, make more deal- purchases, make more purchases in self-service shops or supermarkets, buy more in shops belonging to chains, have more variation in interpurchase times and in volume per purchase occasion and buy more different package sizes.To a certain extent brand choice behavior was found to be transitive over products, i.e. to some extent households showed the same type of brand choice behavior in relation to different products, but not to such a degree that a general brand choice behavior could be spoken of which could serve as an independent basis for market segmentation.
Original language | English |
---|---|
Qualification | Doctor of Philosophy |
Awarding Institution | |
Supervisors/Advisors |
|
Award date | 10 May 1974 |
Place of Publication | Rotterdam |
Publisher | |
DOIs | |
Publication status | Published - 10 May 1974 |
Keywords
- trade marks
- publicity
- advertising
- design
- intellectual property rights
- industry
- property
- policy
- consumers
- households
- consumption
- consumer affairs
- financial management
- cum laude