I. Market Segmentation by Age One important question facing Mikesell s concerns whether millennials differ from non-millennials in various characteristics related to chips and snacking. To get a jump on this issue, we should look at the age data, make some adjustments, and look at some research questions (much as we would hypotheses). The data s measure of age is a little confusing because it frames it by date of birth. I changed it to reflect how old the respondent is, so it s a little easier to understand. Next, we can look at the age cohort of respondents. First, let s get a frequency distribution of ages from the SPSS. What is your age? Frequency Percent Valid Percent Cumulative Percent Under 20 years 1.2.2.2 20 to 37 years 38 6.2 7.8 8.0 Valid 38 to 49 years 130 21.1 26.6 34.6 50 to 68 years 278 45.1 56.9 91.4 69 years or over 42 6.8 8.6 100.0 Total 489 79.4 100.0 Missing 9 127 20.6 Total 616 100.0 Look at the column labeled Valid Percent. This excludes missing values from the percent frequency. Notice that the data are skewed toward those over 50. Notice also that only one person under twenty responded to the survey, and only thirty-nine people under thirty-eight years old responded. Examining millennials will be difficult because there are simply not enough of them compared to older age groups to conduct valid statistical tests. What I did instead was to divide the data into two groups by creating a new variable called age_cohort. This divides ages into two groups. The first are people under fifty years old. These are called the Younger than boomer group. The second is called the Boomer and older group and is comprised of people fifty and over. To analyze data by age, we can approach it two ways. One is to use the age variable as it is to test correlations between age and various chip and snacking related variables. The other is to use the cohort approach and use analysis of variance to model variables according to membership in either of the cohort groups defined by age_cohort. I suggest doing some of both, though in many cases, the results will be similar. However, the cohort analysis will likely be more descriptive for the clients. We can try to use the thirty-nine responses under thirty-eight years old to make some comparisons, but my they will have to be markedly different from other age groups to have the statistical power needed for effective hypothesis testing.
To follow are questions you can use to analyze the data according to age. You really do not need conceptual hypotheses for this analysis, though they certainly would not hurt. However, to answer the questions, you can analyze the data in ways you should already know. Overall Chip Eating Behavior 1. Does age affect chip eating behavior? likelihood of eating snacks frequency of eating snacks occasions for eating snacks places where snacks are purchased 2. Does age affect self-image, positive or negative? 3. Does age affect respondent connectedness? technology use social media use social media search for product and chip information 4. Does age affect brand interactions? chip brand awareness chip purchase likelihood chip brand preference (Mikesell s versus Lay s) brand value perceptions importance factor brand experience importance factor brand occasion importance factor attitude toward chips (first and second measures) Brand Knowledge, Purchase Likelihood, and Price 1. Does age affect awareness of particular brands? Examine by correlation (using the age variable) or analysis of variance (using the age_cohort variable) whether respondent age affects the familiarity toward given brands. 2. Does age affect likelihood of purchase? Again, using either the age or the age _cohort variables, examine the same question about likelihood of purchase of the various brands. 3. Does age affect the maximum willing price? The data include questions about what respondents would be willing to pay for particular brands of chips. Test these for age effects.
II. Market Segmentation by Respondent Sex Now, we can ask many of the same questions regarding sex differences between respondents. To see whether such tests are feasible, we can do a frequency distribution by respondent sex. Are you male or female? Frequency Percent Valid Percent Cumulative Percent Male 221 35.9 45.2 45.2 Valid Female 268 43.5 54.8 100.0 Total 489 79.4 100.0 Missing 9 127 20.6 Total 616 100.0 Looking again at the Valid Percent column, of those reporting their sex, about ten percent more are female than male. Therefore the sample is easily balanced enough for effective hypothesis tests of sex differences on many of the same variables as with age or age cohort. To follow are questions you can use to analyze the data according to respondent sex, which are essentially identical to the questions asked above for age. Overall Chip Eating Behavior 1. Does respondent sex affect chip eating behavior? likelihood of eating snacks frequency of eating snacks occasions for eating snacks places where snacks are purchased 2. Does respondent sex affect self-image, positive or negative? 3. Does respondent sex affect respondent connectedness? technology use social media use social media search for product and chip information 4. Does respondent sex affect brand interactions? chip brand awareness chip purchase likelihood chip brand preference (Mikesell s versus Lay s) brand value perceptions importance factor brand experience importance factor brand occasion importance factor attitude toward chips (first and second measures
Brand Knowledge, Purchase Likelihood, and Price 1. Does respondent sex affect awareness of particular brands? Examine using analysis of variance whether respondent age affects the familiarity toward given brands. 2. Does respondent sex affect likelihood of purchase? Examine the same question about likelihood of purchase of the various brands. 3. Does respondent sex affect the maximum willing price? The data include questions about what respondents would be willing to pay for particular brands of chips. Test these for respondent sex effects. III. Interactions Between Age and Sex Sometimes grouping variables such as cohort and sex can work together to produce more pronounced results in some combinations of groups than others. ANOVA is generally used to test these types of results, which are called interactions. For a few, but not all, of the questions above, possible interactions between cohort and sex may emerge. IV. Market Segmentation by Income The questionnaire included a question about respondent household income, which can be a pretty useful item for segmentation purposes. Unlike the distribution of ages, the distributions of income are as expected, with larger frequencies in the middle income ranges, as shown below. Surprisingly, only five respondents refused the item. What is your combined annual household income? Frequency Percent Valid Percent Cumulative Percent under $20,000 18 2.9 3.9 3.9 20,000-69,999 157 25.5 34.4 38.3 Valid 70,000-119,999 144 23.4 31.5 69.8 120,000-169,999 77 12.5 16.8 86.7 170,000 or more 61 9.9 13.3 100.0 Total 457 74.2 100.0 Missing 9 159 25.8 Total 616 100.0
To consider including income as a segmentation variable, you should analyze the data to answer the following questions. 1. Does income affect respondents overall chip eating behaviors? likelihood of eating snacks frequency of eating snacks occasions for eating snacks places where snacks are purchased 2. Does income affect self-image, positive or negative? 3. Does income affect respondent connectedness? technology use social media use social media search for product and chip information 4. Does income affect brand interactions? chip brand awareness chip purchase likelihood chip brand preference (Mikesell s versus Lay s) brand value perceptions importance factor brand experience importance factor brand occasion importance factor attitude toward chips (first and second measures) V. Segmentation by Household Composition The data include three variables that may be of interest to further examining the composition of our sample for segmentation purposes. First is marital status, a nominal variable indicating simply whether the respondent is single or married. Finer gradations such as widowed or divorced were not asked because they tend to be quite sensitive. Second is number of children under seventeen living at home with the respondents. Values are scaled. Third is a variable I created just below the number of kids variables called one_kid_min and labeled Whether household has at least one child living there. Most respondents live in homes with no kids. Given the age of the sample, my guess is that most parents in the sample have grown kids who no longer live with them. Still, the sample is evenly enough divided to make statistically valid comparisons. Using correlation and ANOVA, examine these variables to see how they relate to chip eating behavior and other chip and brand related variables.
Notes on Race and State of Residence There may be reason to believe that race could be an effective segmentation variable. We all recognize that different ethnic groups have different dietary traditions, which may well extend to snacking. The issue with our data is the distribution of races, which will make statistical analyses difficult. Not including missing values, over 85% of the sample is white. This kind of imbalance can produce invalid or unusual results, particularly with respect to hypotheses testing. Regarding state of residence, while the majority of respondents are clustered in states where Mikesell s is distributed, the distribution among the states is actually pretty good. However, analyzing differences by state or by region will be very cumbersome and more time-consuming than useful.