[meteor_slideshow slideshow=”adssa” metadata=”height: 126, width: 630″]
We will go step by step. Market research is highly important as it helps you understand the context in which you are doing business. According to American Marketing Association dictionary, market research is “the systematic gathering, recording, and analyzing of data with respect to a particular market, where market refers to a specific customer group in a specific geographic area”. On the other hand, marketing research they define as “the function that links the consumer, customer, and public to the marketer through information – information used to: – identify and define marketing opportunities and problems; – generate, refine, and evaluate marketing actions; – monitor marketing performance; and – improve understanding of marketing as a process. Marketing research: – specifies the information required to address these issues, – designs the method for collecting information, – manages and implements the data collection process, – analyzes the results, and – communicates the findings and their implications. Therefore, what you need to do for the final project is exactly what is specified above. Here are some notes, following the required structure of your research project. 1. Research question a. What do you want to research? You need to have in mind the PROBLEM that you want to address. Based on this problem, you need to generate a clear research question. Generally, you could be interested to: describe, understand, relate or observe something. This question needs to be stated clearly: e.g. instead of asking “how to solve problems we are facing?”, ask questions like “what is the key for driving our sales?”, “What is the impact of this ad on consumer’s willingness to buy the advertised product?”, “How do consumers use our products?”. If you have trouble focusing the question, you can analyze the secondary data for several questions (see point 2) and then focus your primary data collection on a clearer question. In the class you did this part by sending me your questions and discussing it during seminars. As soon as you define a research question (or a set of research questions), try to provide a plausible answer and clear arguments why you expect such answer. This will help you see whether your question is reasonable. If you can easily respond to your question – then you might not need the formal research and data collection. For example, if your questions was “What is our market share?”, and you can look at firm’s sales record and conclude that it is 25%, then there is no need to collect data on consumers. However, if you cannot provide a clear answer to your question (e.g. Why are our sales dropping?), then it implies that your question needs further refinement. Have in mind also that your questions (or actually the answer to that question) should have manageable implications. For example, finding an answer to the question “How many people will buy product X?” will give you an idea of the market, but more important is the question “How can we drive the sales of product X?”. Having a question which has managerial relevance turns you (as a market researcher) into “consultant” rather than “data processor”. Once you have clearly defined your research question, reformulate it in terms of “boxes and arrows”. In this context, the “boxes” are actually the variables that you are interested in, and the “arrows” are relationships between those variables. Naming the variables is important. Use the names so that they mean something. Have in mind that you have to measure this variable. For example, calling a variable “advertising” does not mean anything. You should specify what it is about advertising you are interested in. A better variable name would be “advertising frequency” or “advertising likability”, which are variables that mean something and can be measured. Before defining your measures, state a clear definition of your variable. Complete the following sentence “Variable X is defined as _____”. If you cannot define it in one sentence, it means that your variable is not well formulated. When you have a good name and a clear definition which reflects that variable name, then you should decide how to measure it. It can be measured either by using proxy (see secondary data part) or by using questions in a questionnaire (see primary research part). REMEMBER: you generally should NOT measure the link but the variables. Links are established later statistically. b. Why (relevance of the problem)? Your problem needs to be relevant and interesting. In order to have a relevant problem (and related research question) you have to have clear implications (generally managerial ones). Your research results should not just tell what is the level of brand’s reputation, but should also indicate why this reputation is important and how can it be managed. Thus, instead of just providing information, it is important to clarify why is this information relevant. Although, learning the level of a variable is of unquestionable importance (e.g. brand equity), it is only important when the level of this variable has clear implications (either from prior research or from your research results. 2. Secondary data By doing secondary data analysis, you can get an important understanding of what you are interested in. a. What is known on the topic? Analyzing scientific publications (e.g. EBSCO database of journal articles) you can learn how variables are related. For example, there you can find information that loyalty has a positive impact on firm sales. Thus, in your research you do not need to establish this relationship, as it is already well established, but it is only enough to measure loyalty. Similarly, literature might have defined what causes loyalty, thus managerial implications are also relatively clear. Using others’ research helps reduce market research costs as we are not investing in researching links which are well established. b. What existing data can be used to start better understanding the problem? Which sources? Which proxies? Why? Generally, most of the things you can find a secondary data proxy. In many cases information on what you are expecting is relatively easily accessible in secondary data sources (e.g. the census). Moreover, in many industries there are associations in charge of tracking different data. Similarly, many agencies track their consumers continuously (e.g. panels) thus enabling you to use this information to learn about market trends. Defining a good proxy is hard and it requires creativity on your part as a researcher. You need to argue well for your proxy. Remember the example we used in class – how to use secondary data to learn whether a particular person is in love? Good proxies that we found were “city council – check whether they are married”, “facebook – check their relationship status or pictures”, etc. c. Present the data that you can acquire to guide you and provide insights. You should actually present the data that you argue would be good proxies. You can use a very limited sample (e.g. 10) just to have an idea how this data can be indicative of the answer to your research question. 3. Primary research a. Which data do you need to get the answer to your research question? As previously indicated, you now need to define which data you need to collect. This implies that for each variable you should define exact questions that you have used. If the variable is simple (e.g. product design likability), you can simply as a question “how much do you like the design of product X”. However, more complex variables (e.g. satisfaction) implies that consumers can interpret satisfaction differently (e.g. one can be highly satisfied with a hotel although the food was bad, or another one can focus on the food and be completely dissatisfied because of the food). For these, more faceted, variables you need multiple questions. As discussed in class, you can have two types of measures for your variables. First (a) formative measures – where you would measure satisfaction by asking consumers “how satisfied were you w
ith hotel staff”, “room”, “cleanness”, “food”, “accessibility”, etc. Thus, in this case you are asking the questions regarding the extent of satisfaction with different aspects that together form satisfaction with the hotel. Second (b) reflective measures – where you would measure satisfaction with questions like “Were you satisfied with this hotel”, “Would you recommend this hotel to your friends/family”, “Did this hotel meet/exceed your expectations”, etc. Thus, in this case you are asking questions that consumer would naturally agree with if he/she was satisfied, i.e. you are asking questions that reflect consumer’s “state of satisfaction”. Once you have defined the measures for these different variables, you need to assess whether what you believed was forming/reflecting the variable actually is doing so. For this purpose we assess validity and reliability of your measures. We focus only on two basic aspects. First, by running factor analysis you see how answers to the questions you have asked are related to each other. If they are closely related, factor analysis will define them as one factor. Then, it is up to you to define the name of this factor. As factor analysis is discussed in detail in your book, please read the details there. Generally, you need to enter the measures into factor analysis and it will tell you which questions are linked with which ones. Therefore, if you believed that 3 questions are measuring your variable and factor analysis groups them together, that is a first sign that they are actually measuring your variable well. After factor analysis, you should assess the reliability of your measure. To do this you click on ANALYZE – SCALE – RELIABILITY ANALYSIS. Select only the items that are related to one single variable (e.g. S1, S2, S3 which are related to satisfaction) and click ok. If Cronbach’s Alpha is >.7 then you have a reliable scale. What if alpha is less then .7? Now you as a researcher have to decide what to do. Maybe some of the question really is not related to your definition of the variable you are measuring. If so, exclude this question. If all the questions are highly linked (conceptually) to your variable and its definition, then look at factor loadings (from your factor analysis). Try to exclude those as they would generally improve your measure. It means that a particular measure is not really measuring the same thing as other questions. b. Which method would be optimal? Consider benefits and costs and chose the best option. Some of the methods include: Qualitative research, Experimental research, and Survey. Given that we have decided that everyone will do survey, you need to define your approach: whether it will be observation / survey / experiment / secondary sources / technology aided (e.g. eye tracking); and whether you will collect data individually (e.g. Interview) or in groups (e.g. focus groups). Since in most of the projects we are dealing with attitudes (which is often the case in understanding markets), we have discussed different ways you can measure them. Generally, we distinguished between: • Direct Measurement – Structured – which are much more easy to analyze (we generally focus on these in the course) • Yes/No questions • Likert scale – generic question being: „Please indicate the extent to which you agree with the following statements, with 1 being strongly disagree, and 7 strongly agree“, after which you list a set of statements. • Semantic differential – where you have two oposing ideas (e.g. good vs. bad) and respondent needs to select the extent to which he/she believes something is good vs. bad. • Paired comparison – selecting favorable aspect among alternatives or asking respondent to divide 100 points across alternatives based on a certain rule (e.g. how much he/she likes the brand) – Non-structured – where you allow for an open-ended response, e.g. Please describe how do you feel about the product X. Although such questions do not constrain respondent, they are much harder to analyze as they require the researcher to put all the claims by the respondents into numbers that represent them (i.e. it requires the coding part). • Indirect Measurement – Observation – which enables us to see the consumer in his/her natural surrounding. For example the use of IKEA’s CD holder for toast serving which could not be shown in any survey. – Projective techniques – which are needed to uncover underlying information on respondents (remember the example with washing hands after toilet) • Physiological measures – which allow for very precise measure of consumer responses but are costly and it is hard to get consumers to accept such methods. – skin responses – blood pressure – … Also, remember the bias that different question types induce. Each approach has its benefits, but at the same induces some biases. For example, if I want to find out how much each one of you contributed to your group work – I could take one of the four approaches: Rating – asking respondent to assess on a scale from 1 to 7 (7 being most favourable), how valuable each member’s work was for the group’s output. In most of the cases you would grade everyone with 7. Thus, this information does not provide me as a “researcher” with any information. Ranking – asking respondent to rank the group members from the one who contributed the most to the one who contributed the least. Here I am “forcing” you to provide a relative response. However, what if three people worked equally and one was tagging along? Then I can ask you to divide 100 points across different individuals indicating the extent to which they have contributed. Sorting – asking you to sort each other into groups: e.g. those who contributed to the content vs. those who contributed toward execution (e.g. data collection). If I also asked you to divide points in each group, I get more detailed information on who did what (and not just the effort you put into the project). Choice – asking you to choose the one individual who did the least in your project. By doing so, you as a respondent are forced to select the “weakest link”. The same can be done asking you to select the one who did the most on your project. c. How would you sample respondents? We have discussed several sampling approaches: (a) Probability sampling, which is good when you want to have a random sample of a certain population, e.g. a sample of students; (b) Non-probability sampling, which is good when e.g. a problem is too specific for general public, or it is a preliminary research and thus such sampling reduces costs. Moreover, you need to define which contact methods will you apply, and why: e.g. panel, personal interviewing, mail, telephone (CATI, Completely Automated Telephone Survey, etc), web, etc. For all these choices you need to provide arguments why you believe that that is the best way to collect the data. When arguing for your choices have in mind: (a) your research question and variables, and (b) your respondents (e.g. are they able to use internet to reply, can they assess the product’s design if they only see it on a computer and not in real life, etc.). 4. Data collection a. Due to time/money limitations you are unable to collect the data in the optimal manner as described in point 3. Propose an alternative data collection which would reduce time/money demand but still provide adequate input. Thus, although you believe that some type of probability sampling would be the best to use, in your projects, most of you (due to financial reasons and time) have used non-probability, convenience sample. In this part you need to argue, why the method that you have actually used will still be able to provide us information. Remember, if you need to learn about the “arrow” (e.g. how much price influences purchase likelihood), you do not need necessarily the representative sample. The only thing that you need is variation in your variables (i.e. that there are respondents who were exposed to a range of prices from low to high,
and that there are respondents who exhibit different purchase likelihoods: from definitely would not buy to definitely would buy). On the other hand, if you want to know how much people are willing to pay for your coffee, you would really benefit a lot from having a representative sample of your potential buyers. Make sure you discuss different limitations (i.e. biases) that your results will have because you did not use the optimal approach but a “convenient” one. 5. Data analysis a. Which methods you used? Why? A. Clearing the data. In this part you should make sure that your respondents are (a) competent enough to respond to your survey. To ensure this, you should do at least the following three things. First select the respondents who seem “natural” for your question. For example, if your research is about use of food supplements in cooking, students are generally not a got sample. Second, you should ask your questions in a way that your respondents can understand and answer them. For example, if you want to learn something about older citizens who are illiterate, you should adjust your questionnaire using e.g. pictorial evidence. Third, ask a direct question at the end asking respondent the extent to which he/she feels competent for answering the administered questionnaire. In addition, you should make sure that your respondents are (b) motivated enough to respond to your survey. This should be checked by asking similar questions in different parts of your survey to ensure they are concentrated and focused. If the questionnaire is short, this becomes a negligible issue. Third, you should check (c) respondent’s proneness to give high vs. low responses. We will not work with this in the class, but it is important to understand what e.g. number 7 means in respondent’s eyes. “Completely agree” on a 1-7 point scale means rather different thing to different respondent. This you should check by asking some relatively important questions that tend to have “socially acceptable” responses. B. Getting the overview of the data. For getting data overview it is very useful to use diagrams we have discussed (in SPSS, click on GRAPHS – LEGACY DIALOGS): Histogram – which shows you how the data are distributed for each variable. Enter the variable in the field “variable” and check the box “Display normal curve”. Scatter plot – which indicates a general relationship between two variables. Click on “Simple Scatter” and then “define”. Enter the variable that you would have on the X-axis and Y-axis. Then look at the results. They will indicate whether the data are related (i.e. if there is a pattern in the data) or random (i.e. if the pattern seems not to be there). Box-plot – shows you distribution of your variable for different levels of another variable. Select “simple” and then “define”. Enter the variable you are interested in “variable” field variable indicating categories (e.g. gender) into “category axis” field. These diagrams can show you how good is the data you have collected and if there is enough variation in your data to make conclusions. If the data are not normally distributed (and there is no logical explanation for that) then it means you should collect more data. Also, if there is not enough variation (e.g. all your respondents are 24 years old and you want to know how age impacts preference for brand X), then you need to collect more diverse data. In addition, you need to use cross tabs (ANALYZE – DESCRIPTIVE STATISTICS – CROSSTABS) which will provide you with information how your respondents’ responses differ base on their characteristics. In “row(s)” enter the variable based on which you will compare (e.g. gender). In “column(s)” enter the variable whose level you want to analyze (e.g. loyalty). If you want to use more variables (e.g. you want also income in addition to gender), then you enter that variable in “layer”. If you want to compare, you can also ask the SPSS to provide you percentages (click on “cells”, then select which percentages are you interested in: “row”, “column” or “total”). C. Which methods did you actually use. Besides the factor analysis that you used to analyze the measures, and the approaches for getting an overview of the data, you should discuss which methods you have used for analyzing your respondents’ responses. In class we have (or have planned) to cover: Cluster analysis – SPSS combines together respondents who have provided similar answer patterns. Factor Analysis – SPSS combines together questions for which respondents’ answers exhibit similar patterns. Thus, unlike cluster analysis which groups respondents (rows), factor analysis groups questions (columns). In this class we focus only on exploratory factor analysis which implies that from a list of questions in a questionnaire it analyzes which questions tend to go together. If we defined the question well in the beginning the questions that are grouped together should be the ones that relate to the variable we started from. If we are satisfied with the factors that are developed, we should save them as a separate variable. For example, we started from satisfaction variable and defined three questions (S1, S2, S3) to measure this variable. After running factor analysis, it groups S1, S2 and S3 together. Now, we should save it as a variable and name that variable “satisfaction”. Go to ANALYZE – DIMENSION REDUCTION – FACTOR. Select S1, S2 and S3 and transfer them into right box (“variables”). Click on “rotation” and select “varimax”. Click on “scores” and put a check next to “save as variables”. Now go to Variable View in SPSS and find the newly created variable “FAC1_2” (or similar). Instead of that, write “Satisfaction”. In “label” column you will find “REGR factor score 1 for analysis 2”. Delete whatever is written in the label column. t-test – it is a special case of ANOVA used to see how respondents responses are different from e.g. theoretical mean. For example if S1 was “Please indicate the extent to which you agree with the statement: this hotel managed to exceed my expectations” (1-strongly disagree, 7 strongly agree). In this case the theoretical mean is 4 since those who circle 4 neither agree neither disagree with the statement. On a 5-point scale, theoretical mean is 3, and on a 6 point scale, theoretical mean is 3.5. Go to ANALYZE – COMPARE MEANS – ONE-SAMPLE T-TEST. Select the variable S1 and move it to “variables” field. Enter the theoretical mean in “test value” field, i.e. 4 in our case and click “ok”. In the column “mean difference” you will read whether respondents generally believe it is higher or lower (negative sign) than the theoretical mean. In the column “Sig. (2-tailed)” you will see if this difference is significant. If in “sig” column the number is <.05 it means that respondents’ attitude is significantly different from the theoretical mean. If “mean difference” is positive, then it means that respondents generally feel that the hotel exceeded their expectations. If “sig” is >.05 then we conclude that they neither disagree nor agree with the statement (i.e. as if everyone selected 4 as their response). ANOVA – it shows how responses between groups differ. For example if in t-test we see that respondents believe that hotel exceeded their expectations, we want to find out whether men and women in the sample feel the same. Go to ANALYZE – COMPARE MEANS – ONE-WAY ANOVA. In the “dependent list” put variable that you are interested in (in our case S1). In the “factor” put what you want to compare (in our case it is gender). Again look at the “Sig.” column. If it is <.05 it means that responses of the selected groups are different (in our case men and women feel different about the extent to which hotel exceeded their expectations). However, we do not know how they are different but only that they are different. To find out how they differ, we run again ANOVA (s
ame procedure), but before clicking “ok” we select “options” and check next to “descriptive”. Now, the first table in the output shows how men and women responded (the “mean” column) and the lower table shows whether their responses are significantly different. Regression – it is used to show the impact of one variable on another variable. For example, we are interested how their satisfaction impacts their loyalty. To run regression we click on ANALYZE – REGRESSION – LINEAR. In the “dependent” field put the dependent variable, i.e. the one to which the arrow (in the boxes and arrows picture) is pointing to (in our case “loyalty”). In the “independent(s)” enter all the variables which you believe that impact the dependent variable (in our case “satisfaction”. In the table “coefficients” you can read how these variables impact each other. First you look at “standardized coefficients beta”, whether the impact of satisfaction on loyalty is positive or negative. Then, you look at “sig” column to see if the impact is significant (if the “sig” is <.05 it is significant impact). Besides this, looking at the “model summary” table you can see how well you describe your variable, i.e. whether the things you believe impact it actually impact it strongly. “R square” shows you how many % of your loyalty is explained by satisfaction and the “sig” column in table “ANOVA” shows you whether this percentage is significantly greater than zero. If both the beta and the model are significant then you have managed to find an important relationship. b. Which insights can be drawn from the data? This is where you need to be creative. Have in mind that if consumers are loyal 6 on a 1-7 point scale, it has limited information for the firm. It becomes much more important information if we also have information that competitors’ consumers are loyal to them 4 (meaning that we are much better performing) or 7 (meaning that actually we do not have loyal consumers). c. Are there any counter-intuitive ideas in the data? If you find everything that is already well known, your research is not really interesting. However, if you manage to find some counterintuitive ideas, you should focus on explaining why you find such counter intuitive results. Of course, in the sample research that you are doing for this class, I am not expecting you to find some revolutionary new ideas but still try to argue why some of your findings are counter intuitive. 6. Executive summary a. List key findings and recommendations on how to address the research question. b. Clearly argue for your recommendation and argue against other approaches to problem solution. – Explain why the manager should rely on your recommendations rather than on his/her intuition or someone else’s results. c. Which types of research would you like to do as follow up to see the effects of your approach? – This is where you as a researcher are trying to get yourself another “research consultancy job”. It is important that you clearly argue what would be the costs and benefits of this further research. so I chose Nokia company.
[meteor_slideshow slideshow=”best” metadata=”height: 126, width: 630″]