1. [Head-n-Tail Analysis (Part 1): The Toolkit]

    In Part I, I explained what head-n-tail analysis is. Here I will show how you can make an analysis dashboard of your own data in no time.

    To make it simple, I made a dashboard template (on English Premiere League teams TV revenue) for download. All you need to do is to plug in your data and generate a dashboard of your own in 30 seconds.

    Assume that you have your data ready in an excel file with one dimension and one measure. Make sure the data is on "Sheet1". Here is a few data sets for you to try: File1 and File2.

    Then follow these 3 steps: (based on File1)

    1.Edit Data Source (data connection) to load your excel file. The data must be on "Sheet1".


    You may see columns with nulls and  red ! sign. They are the old dimension and measure along with those associated calculated fields. Just ignore them and click "Go to worksheet".



    2.Replace the old dimension "Team" and the old measure "TV Revenue" by those in your data using Replace References. In this case, "Team" is mapped to "Product" and "TV Revenue" is replaced by "Sales".



    3.Edit "Number of Rows" to be that of your table (excluding header row). There are 13 rows in this example.


    Voila, you have your Head-n-Tail analysis dashboard for coffee sales by product.

    You can customize titles, text and colors later at your convenience. Individual charts and worksheets are available for you to add to other dashboards if needed. And your data source can be any database, not limited to excel files.

    Give it a try and see if you can beat 30 seconds.


    2

    View comments

  2. [Head-n-Tail Analysis (Part 2): 30 sec to Dashboard]

    What is head-n-tail analysis? It consists of top ranking and tail aggregation.

    Recently I have had a few different analysis cases which can all use the same simple yet powerful methodology. It seems quite a fundamental analysis that can be applied widely.

    Given a group of salesmen, stores or league teams and the sales, profit or revenue they each made, we are always interested in knowing
    - the top performers (Head)
    - the performance of the rest (Tail)

    For the tail, we will just give the aggregate result. And I would call it Head-'n'-Tail analysis. (Let me know if you can suggest a better name) It is a mix of both ranking and aggregation. The interesting part of this analysis is, it not only gives us the top performers but also the aggregation view of the tail.

    Similar to Pareto analysis, it's important to know that 20% of the salesmen make up 80% of the sales. It's just as interesting to know that the other 80% of salesmen make only 20% of the sales. Moreover, this Head & Tail analysis tells who the top guys are.

    I will present here a simple and quick toolkit for the analysis. I assume that the data set has got 2 columns: one dimension and one measure.

    The Head-n-Tail analysis toolkit consists of 3 elements, all having both head and tail:
    - Ranking
    - Pareto analysis
    - Pie chart

    The tail size is adjustable to be short or long.

    The difference the toolkit makes relative to traditional approach is
    - Ranking includes the comparison of top guys and the aggregation view of the rest of the group.
    - It keeps the running percentage details for the top guys and aggregates the rest together
    - Pie chart used to be messy with too many slices. The new one will only show a few if you wish.
    - A knob is provided for you to adjust the size of head or tail for your analytic pleasure.

    As an example, we analyze the TV revenue for the English Premiere League teams in 2012-13.  From the following charts, we see
    - the top 5 teams and their respective revue (in million euros), and the aggregate revenue of the rest.
    - the running % of total league revenue of the top teams and the rest
    - the visual pie share of top teams and the rest



    In the next part, I will show how you can create a Head-n-Tail analysis dashboard in 30 seconds, based on the above toolkit.



    0

    Add a comment

  3. The color palette we use the most in Tableau is either discrete or gradient. Discrete colors are applied to categorical data or dimensions. Gradient colors are applied to numerical data or measures.

    Here we try to go beyond the dichotomy of discrete and gradient. Rather our approach is a combination of both discrete and gradient color palette. We will use color to categorize numerical data and apply gradient color to a subset of the data.

    Below I will show a few special ways to color your data, simply using the color editor in Tableau. The purpose is to create big contrast between data and to accentuate the difference in the data of interest.

    To create the maximum contrast, we always try to use the following:
    -diverging color palette
    -full color range
    -setting center, start and end values.

    The choice of Center, Start and End values must be judicious.

    Example 1. Differentiate zero and positive numbers

    At one time, I wanted to create a regular heat map of dynamic data, which are mostly positive integers. There are occasionally zeros that people need to be alerted about.

    Then I select a red-green palette and set Center=1 Look at what I got:

    So you see it's quite simple to create 2 categories of data using different colors: zero and positive numbers. And the latter is in gradient color.

    Example 2. Differentiate numbers below and above a threshold

    This is used when we care about the grades at one side of the threshold and get alerted when the values go to the other side. For example, we paint those numbers above threshold in red and those below the threshold in gradient green.

    In the example below, we deal with an upper threshold = 75. We only care about those numbers between 0-74. Anything that is 75 and above is paint red.

    Here we set Center=74.5 and End=75. Thus any number above 75 will be in full red. From 0 to 74, it will be in gradient green.

    Note that the color is reversed in the editor: green for lower numbers and red for above 75.

    Obviously, this approach also works for lower threshold.

    Example 3: Make a bi-color map
    In case where data is split into two categories: above threshold and below threshold, we need to apply only two colors to the data. No gradient color is needed. Say, for data <=15 it's red. For data >=16 it's green. This can be done by setting center=15.5, start=15 and end=16.

    The above could be done using calculated field as well. But I believe using color editor is a simpler approach.

    The workbook is available here for further exploration.
                                                                                   


    0

    Add a comment

Blog Archive
Loading
Dynamic Views theme. Powered by Blogger. Report Abuse.