Segmentation-3

Dendrogram. The dendrogram is a tree diagram to illustrate the arrangement of clusters produce Scree plot The screeplot displays, for each cluster solution, a measure of within-cluster heterogene observations that are widely different (which will happen if the number of clusters is too variability in the data), the value will be high. A good cluster solution might be where the screeplot displays an 'elbow', that is, where i clusters beyond a certain point does not dramatically decreases within-cluster heterogen The measure displayed in the screeplot is related, but not equivalent, to the distance rep

ed by hierarchical clustering, and how the observations are incrementally clustered together. eity. If clusters group o small to capture the increasing the number of neity. ported in the dendrogram.

Segment differences per segment. Cell colors indicate to what extent a segment is statistically diff Segmentation space The chart below is a graphical representation of the various segments, segment member is obtained by plotting the first two dimensions of a principal component analysis perform segmentation data, on top of which segment information has been overlaid. Because only the first two dimensions of the PCA are displayed, and these two dimensio variance in the data, some differences between segments might not appear here. Note t variance, if any, have been excluded. Two clusters that appear to overlap on the first two dimensions might be distinct on othe this chart is a useful guide, for checking which variables are correlated, but may be misle optimal number of segments.

58 1 3 59 1 4 60 3 2 61 2 2 62 3 3 63 2 2 64 2 2 65 2 3 66 1 4 67 3 2 68 3 1 69 1 2 70 3 1 71 2 2 72 2 2 73 3 1 74 3 3 75 1 2 76 1 4 77 2 1 78 3 3 79 2 2 80 2 4 81 1 2 82 1 3 83 3 1 84 1 2 85 2 3 86 1 3 87 1 1 88 1 2 89 2 1 90 1 4 91 3 4 92 2 3 93 3 1 94 3 2 95 1 4 96 2 3 97 3 3 98 1 2 99 2 2 100 2 3 101 1 4 102 2 4

193 3 3 194 3 2 195 2 3 196 3 3 197 3 3 198 2 2 199 3 1 200 3 1 201 2 1 202 3 1 203 1 3 204 3 1 205 2 2 206 3 4 207 3 3 208 2 2 209 3 1 210 3 2 211 2 4 212 3 2 213 3 1 214 2 1 215 3 1 216 2 3 217 3 2 218 3 1 219 2 4 220 2 3 221 2 3

component analysis. only available in the Excel formatted output.

Segment Age 1 1 4 3 1 3 5 4 1 3 17 6 1 3 30 8 1 3 17 9 1 2 11 1 3 12 1 4 16 1 4 17 1 4 19 1 2 21 1 2 22 1 3 30 1 3 31 1 2 32 1 2 33 1 4 34 1 3 35 1 3 40 1 3 41 1 4 47 1 3 48 1 4 50 1 1 53 1 2 56 1 3 57 1 4 58 1 3 59 1 4 66 1 4 69 1 2 75 1 2 76 1 4 81 1 2 82 1 3 84 1 2 86 1 3 87 1 1 88 1 2 90 1 4 95 1 4 98 1 2 101 1 4 104 1 2

109 1 3 112 1 2 116 1 4 118 1 3 138 1 3 142 1 2 143 1 2 146 1 2 148 1 1 149 1 3 152 1 2 167 1 2 170 1 4 176 1 2 178 1 2 179 1 4 203 1 3 2 2 2 7 2 2 14 2 1 15 2 4 20 2 2 25 2 3 26 2 2 27 2 4 28 2 2 29 2 4 37 2 3 38 2 2 39 2 3 42 2 3 44 2 3 46 2 2 49 2 2 51 2 3 52 2 2 54 2 1 61 2 2 63 2 2 64 2 2 65 2 3 71 2 2 72 2 2 77 2 1 79 2 2

80 2 4 85 2 3 89 2 1 92 2 3 96 2 3 99 2 2 100 2 3 102 2 4 103 2 3 108 2 2 110 2 1 111 2 1 114 2 1 115 2 4 121 2 2 122 2 2 124 2 4 127 2 2 128 2 2 129 2 3 130 2 4 132 2 4 133 2 2 134 2 3 135 2 2 136 2 1 137 2 2 139 2 4 141 2 2 144 2 3 147 2 2 150 2 2 151 2 3 156 2 2 157 2 2 158 2 2 159 2 3 160 2 2 161 2 1 162 2 2 171 2 1 172 2 2 175 2 2 180 2 4 182 2 2

184 2 3 186 2 2 187 2 3 189 2 2 191 2 2 195 2 3 198 2 2 201 2 1 205 2 2 208 2 2 211 2 4 214 2 1 216 2 3 219 2 4 220 2 3 221 2 3 5 3 3 10 3 2 13 3 3 18 3 2 23 3 4 24 3 4 36 3 2 43 3 2 45 3 3 55 3 2 60 3 2 62 3 3 67 3 2 68 3 1 70 3 1 73 3 1 74 3 3 78 3 3 83 3 1 91 3 4 93 3 1 94 3 2 97 3 3 105 3 1 106 3 3 107 3 2 113 3 2 117 3 1 119 3 4

120 3 2 123 3 3 125 3 2 126 3 2 131 3 2 140 3 2 145 3 2 153 3 3 154 3 2 155 3 3 163 3 2 164 3 2 165 3 2 166 3 2 168 3 2 169 3 3 173 3 3 174 3 2 177 3 3 181 3 2 183 3 2 185 3 2 188 3 1 190 3 1 192 3 2 193 3 3 194 3 2 196 3 3 197 3 3 199 3 1 200 3 1 202 3 1 204 3 1 206 3 4 207 3 3 209 3 1 210 3 2 212 3 2 213 3 1 215 3 1 217 3 2 218 3 1

Spider chart Spider chart comparing the averages of the segmentation variables across all segments. Spider chart. Segment 1 profile The following charts represent the profile of each segment. These charts are only availab standardized, hence the model assumes that all segmentation variables use the same sca - For each segment, the segmentation variables are ordered in decreasing order of mag - The colored dots represent the average of the segment.

Segment 1 profile. Segment 2 profile - The horizontal lines represent the standard deviations within that segment. - The vertical, gray lines represent the averages of the rest of the population, after exclu

Segment 2 profile. Segment 3 profile

Segment 3 profile.

ble when the data are not ale. gnitude.

uding members of the segment under scrutiny.

Descriptors This table reports the descriptor averages of each segment. The more differences can be predict segment membership based on descriptors alone. Descriptor data per segment Average value of each descriptor, overall and within each cluster. Descriptors that are statistically Population Segment 1 Fst2Try 2.67 2.16 Fashnista 3.34 2.69 PplThnkMe 3.63 3.05 PplAskMe 3.35 2.54 FashnMzne 3.93 2.75 NoBgDeal 2.95 3.30 NotBld 4.29 4.38 OnlyBest 5.12 4.72 ExclsvShop 3.12 2.07 WhrFrndShop 2.41 1.56 OtherPpl 3.72 2.64 MagznNews 3.90 3.15 Catlg 4.48 3.75 Internet 2.25 1.62 Excitbl 4.16 4.39 Thrfty 4.00 3.57 Rationl 3.50 3.03 Yuthfl 4.34 4.80 Orthdx 5.04 5.03 Modst 3.32 3.00 Age 2.45 2.84 ApparelEx 2.03 1.64 Income 2.70 2.54 Edu 1.81 1.78

Descriptor differences per segment. Cell colors indicate to what extent the distribution of a descr Descriptor space The chart below is a graphical representation of the various segments, segment member by outputing the first two dimensions of a principal component analysis performed on th top of which segment information has been overlayed. Because only the first two dimensions of the PCA are displayed, and these two dimensio variance in the data, some differences between segments might not appear here. Note t if any, have been excluded. If two or more segments fully overlap, it is unlikely that they could be clearly separated b However, two segments that seem to overlap on two dimensions may be more clearly se Consequently, the confusion matrix is a better guide to assess the quality of segment cla descriptors.

Descriptor space. Spatial representation of segments and their descriptors, using principal compo

e found, the easier it will be to y different from the rest of the population are highlighted in red (lower) or green (higher). Segment 2 Segment 3 2.85 2.89 3.52 3.68 3.79 3.94 3.62 3.70 4.29 4.48 2.92 2.69 4.30 4.21 5.16 5.42 3.35 3.73 2.80 2.66 4.12 4.15 4.43 3.90 4.81 4.68 2.66 2.27 4.17 3.96 3.91 4.49 3.70 3.66 4.18 4.15 4.87 5.27 3.36 3.55 2.42 2.15 2.03 2.37 2.74 2.79 1.78 1.87

riptor in a segment is statistically different from the rest of the population. rs, and descriptors. It is obtained he (standardized) descriptors, on ons capture only 31.2% of the that descriptors with no variance, based on descriptors alone. eparated on other dimensions. assification based on

onent analysis.

Introduction Often, segmentation (needs) variables for each customer may not be available to manag customers may be available. In this section, we explore whether descriptors alone can predict segment membership w confusion matrix and hit rates (reported below) indicate whether the model is accurate e For member classification based on descriptors, Enginius uses a multinomial logit model predict 'choices between multiple alternatives (A/B/C)' in the predictive modeling modul The largest segment is selected as the default option (dummy), and the model identifies significant for predicting cluster memberships. If a descriptor is highly predictive, its p-va zero, and the cells will appear in green (or red). Model coefficients Model parameters Segment 2 is the model baseline. Segment 1 Segment 3 (Intercept) 5.50 -2.48 Fst2Try -0.024 -0.153 Fashnista -0.012 0.044 PplThnkMe 0.077 0.141 PplAskMe 0.078 -0.201 FashnMzne -0.376 0.185 NoBgDeal 0.106 -0.115 NotBld -0.047 0.087 OnlyBest -0.012 0.164 ExclsvShop -0.290 0.089 WhrFrndShop -0.364 -0.102 OtherPpl -0.265 0.110 MagznNews -0.087 -0.502 Catlg 0.029 0.228 Internet -0.343 -0.202 Excitbl -0.202 0.031 Thrfty 0.020 0.302 Rationl -0.278 -0.173 Yuthfl 0.090 0.021 Orthdx 0.193 0.299 Modst -0.133 -0.039 Age -0.056 -0.538 ApparelEx -0.405 0.241 Income -0.018 0.170 Edu 0.023 0.017

P-values p-values Probabilities that parameter estimates are different from zero only by chance. Segment 1 Segment 3 (Intercept) 0.026 0.218 Fst2Try 0.913 0.396 Fashnista 0.952 0.815 PplThnkMe 0.735 0.488 PplAskMe 0.712 0.245 FashnMzne 0.026 0.187 NoBgDeal 0.544 0.423 NotBld 0.776 0.545 OnlyBest 0.937 0.247 ExclsvShop 0.061 0.471 WhrFrndShop 0.112 0.490 OtherPpl 0.083 0.391 MagznNews 0.670 0.005 Catlg 0.878 0.196 Internet 0.072 0.125 Excitbl 0.172 0.815 Thrfty 0.896 0.055 Rationl 0.039 0.140 Yuthfl 0.555 0.873 Orthdx 0.307 0.054 Modst 0.461 0.807 Age 0.838 0.032 ApparelEx 0.232 0.312 Income 0.915 0.194 Edu 0.943 0.946 Confusion matrix The confusion matrix compares actual segment membership (obtained from the segmen segmentation variables) and predicted segment membership (obtained from the in-samp descriptors alone). When actual and predicted segment memberships coincide, the diago large, indicating that the classification model based on available descriptors is accurate. Confusion matrix (count) The model has correctly classified 143 of the 221 observations. The off-diagonal elements are cla Predicted 1 Predicted 2 Segment 1 42 9 Segment 2 14 62 Segment 3 9 23 Total 65 94

Confusion matrix (%) The global hit rate of the model is 65%. The diagonal elements represent segment-specific hit rat Predicted 1 Predicted 2 Segment 1 69% 15% Segment 2 16% 70% Segment 3 13% 32% Confusion matrix (plot). Graphic representation of the confusion matrix: actual segment member Model predictions

Model predictions (in-sample) This table details the probabilities of each member of the segmentation dataset to belong to each Prob(cluster 1) Prob(cluster 2) 1 72% 26% 2 62% 30% 3 96% 4% 4 58% 40% 5 39% 48% 6 94% 4% 7 4% 84% 8 80% 13% 9 66% 22% 10 18% 31% 11 64% 30% 12 70% 25% 13 0% 34% 14 2% 44% 15 41% 52% 16 82% 16% 17 64% 30% 18 2% 33% 19 55% 27% 20 36% 37% 21 49% 17% 22 53% 19% 23 7% 68% 24 57% 32% 25 64% 31% 26 25% 55% 27 33% 57% 28 14% 26% 29 68% 27% 30 71% 19% 31 28% 36% 32 7% 77% 33 59% 40% 34 78% 17% 35 79% 15% 36 63% 23% 37 40% 29% 38 3% 32% 39 44% 27% 40 11% 69% 41 63% 35%

42 3% 88% 43 9% 25% 44 7% 62% 45 15% 41% 46 9% 45% 47 78% 14% 48 82% 14% 49 10% 62% 50 35% 48% 51 9% 47% 52 2% 54% 53 19% 51% 54 12% 32% 55 41% 42% 56 20% 53% 57 79% 17% 58 88% 5% 59 24% 11% 60 8% 58% 61 7% 50% 62 60% 15% 63 10% 47% 64 2% 56% 65 17% 73% 66 94% 5% 67 1% 35% 68 4% 43% 69 46% 22% 70 3% 48% 71 4% 62% 72 16% 44% 73 43% 38% 74 5% 52% 75 62% 30% 76 61% 33% 77 63% 28% 78 30% 19% 79 6% 64% 80 58% 39% 81 79% 17% 82 91% 5% 83 12% 77% 84 18% 51% 85 14% 73% 86 93% 5%

87 48% 27% 88 7% 40% 89 0% 33% 90 24% 32% 91 15% 68% 92 7% 60% 93 0% 60% 94 4% 69% 95 78% 9% 96 3% 52% 97 4% 41% 98 47% 18% 99 30% 25% 100 47% 24% 101 84% 11% 102 34% 43% 103 4% 84% 104 31% 28% 105 26% 39% 106 49% 24% 107 11% 50% 108 30% 47% 109 79% 20% 110 6% 40% 111 28% 46% 112 5% 26% 113 2% 28% 114 27% 57% 115 4% 93% 116 35% 27% 117 3% 57% 118 48% 47% 119 30% 14% 120 48% 36% 121 2% 69% 122 11% 76% 123 56% 7% 124 80% 15% 125 2% 11% 126 5% 34% 127 41% 25% 128 2% 70% 129 4% 66% 130 2% 59% 131 0% 37%

132 67% 17% 133 24% 41% 134 2% 53% 135 1% 74% 136 0% 42% 137 12% 55% 138 12% 75% 139 21% 59% 140 33% 10% 141 5% 51% 142 65% 9% 143 96% 2% 144 68% 28% 145 0% 22% 146 43% 35% 147 12% 72% 148 14% 50% 149 85% 6% 150 2% 54% 151 6% 78% 152 34% 55% 153 3% 87% 154 9% 70% 155 58% 17% 156 21% 60% 157 7% 68% 158 3% 67% 159 12% 67% 160 1% 82% 161 1% 71% 162 2% 89% 163 1% 41% 164 1% 41% 165 21% 45% 166 6% 46% 167 94% 2% 168 21% 53% 169 8% 55% 170 88% 8% 171 1% 22% 172 77% 12% 173 0% 67% 174 27% 42% 175 20% 47% 176 48% 41%

177 30% 54% 178 17% 26% 179 29% 21% 180 14% 74% 181 21% 32% 182 13% 64% 183 2% 30% 184 13% 31% 185 2% 37% 186 30% 47% 187 23% 41% 188 1% 8% 189 2% 61% 190 3% 25% 191 5% 31% 192 2% 35% 193 1% 19% 194 0% 27% 195 1% 81% 196 13% 45% 197 66% 19% 198 11% 54% 199 2% 29% 200 1% 5% 201 2% 63% 202 1% 25% 203 9% 22% 204 14% 57% 205 6% 43% 206 29% 33% 207 3% 36% 208 40% 20% 209 1% 2% 210 1% 53% 211 9% 51% 212 13% 40% 213 0% 29% 214 3% 76% 215 3% 6% 216 9% 65% 217 20% 27% 218 5% 18% 219 2% 82% 220 16% 63% 221 1% 75%

gers, but descriptors variables for with sufficient accuracy. The enough. (similar to the one used to le. which descriptors are the most alues will be close to

ntation analysis and the original ple classification analysis and the onal elements will be comparatively assification errors. Predicted 3 Total 10 61 13 89 39 71 62 221

tes. Predicted 3 Total 16% 100% 15% 100% 55% 100% rship versus predicted segment membership. Bubbles in the diagonale represent correct classification.

h cluster (as predicted by the in-sample classification model and the descriptors alone). The segment with the highest proba Prob(cluster 3) Predicted Actual Correct 2% 1 1 1 8% 1 2 0 1% 1 1 1 2% 1 1 1 13% 2 3 0 3% 1 1 1 12% 2 2 1 8% 1 1 1 12% 1 1 1 51% 3 3 1 7% 1 1 1 5% 1 1 1 66% 3 3 1 54% 3 2 0 7% 2 2 1 2% 1 1 1 6% 1 1 1 65% 3 3 1 18% 1 1 1 27% 2 2 1 34% 1 1 1 28% 1 1 1 24% 2 3 0 11% 1 3 0 6% 1 2 0 20% 2 2 1 9% 2 2 1 60% 3 2 0 4% 1 2 0 9% 1 1 1 37% 3 1 0 16% 2 1 0 1% 1 1 1 5% 1 1 1 5% 1 1 1 13% 1 3 0 31% 1 2 0 65% 3 2 0 28% 1 2 0 21% 2 1 0 1% 1 1 1

9% 2 2 1 66% 3 3 1 31% 2 2 1 44% 3 3 1 46% 3 2 0 7% 1 1 1 4% 1 1 1 28% 2 2 1 17% 2 1 0 44% 2 2 1 45% 2 2 1 30% 2 1 0 56% 3 2 0 17% 2 3 0 27% 2 1 0 4% 1 1 1 7% 1 1 1 64% 3 1 0 34% 2 3 0 43% 2 2 1 26% 1 3 0 43% 2 2 1 42% 2 2 1 10% 2 2 1 1% 1 1 1 64% 3 3 1 53% 3 3 1 32% 1 1 1 49% 3 3 1 35% 2 2 1 40% 2 2 1 18% 1 3 0 43% 2 3 0 8% 1 1 1 6% 1 1 1 9% 1 2 0 51% 3 3 1 29% 2 2 1 3% 1 2 0 4% 1 1 1 3% 1 1 1 12% 2 3 0 31% 2 1 0 12% 2 2 1 2% 1 1 1

26% 1 1 1 52% 3 1 0 67% 3 2 0 44% 3 1 0 16% 2 3 0 33% 2 2 1 39% 2 3 0 26% 2 3 0 13% 1 1 1 45% 2 2 1 55% 3 3 1 35% 1 1 1 44% 3 2 0 29% 1 2 0 5% 1 1 1 23% 2 2 1 13% 2 2 1 41% 3 1 0 35% 2 3 0 28% 1 3 0 39% 2 3 0 23% 2 2 1 1% 1 1 1 53% 3 2 0 26% 2 2 1 69% 3 1 0 70% 3 3 1 16% 2 2 1 3% 2 2 1 38% 3 1 0 40% 2 3 0 5% 1 1 1 56% 3 3 1 16% 1 3 0 29% 2 2 1 13% 2 2 1 37% 1 3 0 5% 1 2 0 87% 3 3 1 61% 3 3 1 34% 1 2 0 28% 2 2 1 30% 2 2 1 39% 2 2 1 63% 3 3 1

16% 1 2 0 35% 2 2 1 45% 2 2 1 24% 2 2 1 57% 3 2 0 33% 2 2 1 14% 2 1 0 20% 2 2 1 56% 3 3 1 44% 2 2 1 26% 1 1 1 1% 1 1 1 4% 1 2 0 78% 3 3 1 22% 1 1 1 16% 2 2 1 36% 2 1 0 9% 1 1 1 44% 2 2 1 17% 2 2 1 11% 2 1 0 10% 2 3 0 21% 2 3 0 25% 1 3 0 18% 2 2 1 25% 2 2 1 30% 2 2 1 21% 2 2 1 17% 2 2 1 28% 2 2 1 10% 2 2 1 58% 3 3 1 58% 3 3 1 33% 2 3 0 48% 3 3 1 4% 1 1 1 26% 2 3 0 37% 2 3 0 4% 1 1 1 77% 3 2 0 12% 1 2 0 32% 2 3 0 31% 2 3 0 33% 2 2 1 12% 1 1 1

17% 2 3 0 56% 3 1 0 50% 3 1 0 12% 2 2 1 47% 3 3 1 23% 2 2 1 69% 3 3 1 56% 3 2 0 61% 3 3 1 23% 2 2 1 37% 2 2 1 91% 3 3 1 37% 2 2 1 72% 3 3 1 64% 3 2 0 62% 3 3 1 80% 3 3 1 73% 3 3 1 17% 2 2 1 41% 2 3 0 15% 1 3 0 35% 2 2 1 69% 3 3 1 94% 3 3 1 35% 2 2 1 74% 3 3 1 69% 3 1 0 28% 2 3 0 51% 3 2 0 38% 3 3 1 61% 3 3 1 40% 1 2 0 97% 3 3 1 46% 2 3 0 40% 2 2 1 47% 3 3 1 71% 3 3 1 22% 2 2 1 91% 3 3 1 26% 2 2 1 53% 3 3 1 77% 3 3 1 16% 2 2 1 20% 2 2 1 25% 2 2 1

ability is retained, and is compared to the actual segment membership to measure model accuracy and classification errors.

.