
Part 2. MM4XL Tools > 2. Analytical Tools > Variation Analyst > Anatomy of a Variation Analyst Output Report Group Variation AnalystAnatomy of a Variation Analyst Output ReportThe tables and charts Variation Analyst prints in output always look the same, although some formulae change in the way they are computed depending on whether the analysis is made with equal or unequal sample sizes. The user is not required to take any action in either case, because the tool can detect automatically which instance applies. To illustrate the output report we use the data of the sheet Data  Equal sample in the Example file that can be opened from the tool form.
There are three groups in the data set: Dash, Dixan and Persil. The values refer to sales growth performance in six geographic regions where Dash ran a promotional action targeting Dixan. The manager wants to know whether the action produced the expected result. Persil was included in the study as a control group. The critical output section of the Variation test is the Group Comparison Report, shown below. Column A contains the pairs of groups, products in our examples, tested for the significance of differences.
Each row shows the result of one test, and there are n(n1)/2 tests run in an analysis, in our case 3 tests = 3*2/2. Column B is the key place to look at. Yes in cell B27 stands for Yes, results measured for Dash are statistically different from those measured for Dixan. The column Desired Probability indicates the probability level employed for the test, 95%. To its right, Achieved Probability indicates the probability level the analysis found for each comparison of group pairs. When the achieved probability is lower than the desired one we must conclude that the variation in the data is not significant and it must be due to casual effect rather than to our promotional activity. Therefore, in these cases column B will display No. FValue and Fcritical contain the base values used by the analysis of variance method to test its assumptions, and for which we refer the reader to any statistics book dealing with this wellknown analytical method. Read also the section Technicalities for more information concerning the Fvalue. Correlation shows the measure of association concerning the input data of the grouppair. The correlation coefficient ranges from 1 to 1. Our example shows a positive association for Dash and Dixan. With reference to the input data, this means when sales of Dash grow, sales of Dixan also grow, and vice versa. On the other hand, in cell G28 we find a negative association between Dash and Persil, meaning that when sales of Dash grow those of Persil shrink, and vice versa. These facts may be due to the peculiar distribution system of the market or other reasons, and in both cases they contribute an extra piece of information useful to frame the context in an appropriate manner. In the Average Difference column we find the overall difference in mean value for the pair in analysis. The value 0.02 in cell H27 stands for an average +2% sales growth for Dash against Dixan. A negative value would have meant that Dixan had performed better in sales growth than Dash did. This value may be used as a reference point for the measurement of the effectiveness of promotional activities (we remind the reader that promotions are strategic activities offered for a short period of time and aimed at increasing sales volume). The Statistics table shows the most common descriptive measures for each group. This information can help us understand the shape of the data we are working with, especially Mean and Variance values as we shall see.
The chart Input Data is a common line chart drawn with the original values as input by the user.
The chart Average Variance plots these two measures for each group, and it can be very helpful to figure out what kind of products (groups) we are dealing with. According to the input data of the example, a high average value such as for Dash may be obtained thanks to a few abnormally large observations only, which wouldnt necessarily indicate an overall positive impact of the promotion. For this reason, the variance is plotted together with the mean. When the variance is high, such as for Dixan, we may conclude there are spread observations in the data ranging from very high to very low (indeed Dixan has the lowest Min value of the three groups). Alternatively, when the variance is low we may conclude most observations are placed around the mean value, which in such a case may be seen as a good estimator of the final promotion outcome (like Persil, which has the smallest value, 0.026 or 2.6%, for the difference MaxMin).
The Quadrant Analysisis the last output element produced by Variation Analyst, and it can prove very useful when dealing with groups made of many rows, such as sales growth values concerning 200 microzones of a whole country, for example. But it may also be very helpful with a smaller number of items, as we are going to see.
Each item (geographic areas in our example) of the input data for a selected pair of groups is assigned to one of four categories. The labels in column A of the summary table refer to the quadrant position in the chart. TopRight, for instance, is the quadrant of the chart hosting zones scoring high on the second as well as the first product. In our example, there are two such zones (2 and 3) for the pair Dash (product 1) versus Dixan (product 2). For the product target of the analysis, Dash, the input data for the two zones Sum up to 5.9% (cell C34) and they have an Average value equal to 3% (cell D34). On average, items belonging to the TopRight quadrant contribute +1.9% (cell E34) as compared to the overall product average value of 2.1% (cell E37). Looking at the quadrant chart may make it easier to understand the concept. The chart is a scatter plot with on the xaxis the input values for the first product of the pair in analysis. In our example this is Dash, and the length of the xaxis goes from 01% to 3.5%, respectively the min and max value for Dash input data (see cells B17 and B18 of the Statistics table). Similarly, the yaxis ranges, in this example, from 1.5% to 2.5%, which are the min and max values for Dixan (cells C17 and C18 of the Statistics table). For both axes the crossing point is set to the midpoint, which equals (MaxMin)/2. This way the chart splits in four quadrants of equal size. Each quadrant has a meaning, and for ease of interpretation they have been labeled as follows:
1. TopLeft: Disappoint These are items where product 1 scored low and product 2 scored high. If we were analyzing the result of a promotional action, areas placed in this region of the chart had reacted badly to the offer of the target product. Typically, there shouldnt be items in this quadrant. 2. TopRight: Head to head Items in this quadrant scored well on both products, which may underline a not very effective promotion on our part. Typically, we should see a reduction in performance of the target competitor as opposed to growth for the challenger one. When both grow one may suspect the growth is due also to reasons independent of the promotion itself. Financially speaking, items in this quadrant are acceptable, but on the strategic side they are not. There shouldnt be too many items here. 3. BottomLeft: Tough jobs These are items where both products scored low. Winning space in these areas seems to be very difficult. In order to have an effective action there should not be many items in this quadrant, although it is quite common to have areas where the performance is lower than in others. The position is sustainable as long as the xaxis doesnt sink into negative values. 4. BottomRight: Got it The optimal outcome of a promotional action is that our product sales grow and those of the target competitor decrease, indicating that we drew market share away from the competitor. This is captured in the lowest quadrant on the right of the chart. When running a promotion, we want to find as many items as possible in this quadrant. A note of caution is required. So far we have been discussing an example based on sales growth data. Some of the statements made above might not hold when dealing with data sets where, for instance, negative values are interpreted as a positive rather than a negative outcome, for instance, reduction in time or savings on costs. 