• http://searchmarketingcommunications.com Tim Cohn

    Pretty cool…

    Haven’t you shown us the data in seven dimensions?

  • http://www.efrontier.com sidshah

    You are right. I have shown data in 7 dimensions. 6 is a typo. Hmm… now we could add a moustache as an 8th. Fu Manchu’s here we come !!

  • http://www.chadsummerhill.com ChadSummerhill

    Very interesting, Siddharth! Your article got me thinking about how cool it would be if Tableau added Chernoff faces to their visualization software. I even added a post to my blog in honor of your article.


  • http://www.efrontier.com sidshah

    Hi Chad,
    Glad you like it. I did the visualization in R. It has a built in library that can do chernoff faces. If you know R its quite straight forward !

    If you speak to hard core data/stats guys they would not like chernoff faces as it has such a qualitative feel. Then again, they are not marketers !


  • http://www.chadsummerhill.com ChadSummerhill

    Hi Sid,

    I wonder how hard it would be to add Size and Color to the R library. A little color could help a lot with this type of visual.

    Ever used Tableau? Written any articles about it?


  • http://www.fingo.co.uk BenMorel

    Wouldn’t this data be more useful in correlation graphs? I have a feeling that othes would be more receptive to it that way.

    Pulling together a quick scatter graph in Excel would instantly show you that ad position and CPA, your chosen reference metric, were not correlated – this column could then be removed from the study as it is not an influencing factor. CPA and conversion rate you would expect to be well correlated and again a quick scatter graph would probably show you that they are – proving quantitatively that your first bullet under “what do these faces tell us?” is correct.

    The biggest things I imagine that Management, or pretty much anyone, would respond to would be correlations of “Visits vs CPC” and “Visits vs conversion rate” – especially if there were any anomolies. The anomoly of Boston’s low conversion rate, for example, would be immediately obvious and all of this data would be actionable.

    How would you action the data presented by the faces? Unfortunately I fail to see the time or visualisation advantage the Chernoff faces bring.

  • http://www.efrontier.com sidshah

    Hi Ben,

    Like I said before quants dont like Chernoff faces . In fact you can find numerous critiques on it. However, I can see instances where they would be very useful

    You can create a correlation matrix with the above data but my personal experience showing to data with non-quant management people is that it can often overwhelm them. This is specially true when you have more than 4 -5 metrics. Also, when you have so many metrics a 2 dimensional correlation often doesn’t tell the story… but thats a different blog post all together :)

    Chernoff faces work well with people who are not oriented to think quantitatively and also when you want to mix disparate data into one qualitative metric. So lets say you want to assess the quality of management of a google campaign. The factors could be

    a. Avg Quality score
    b. Avg CTR
    c. Avg. number of ad variations
    d. Volume
    e. Ratio of keywords to ad groups
    f. Total keywords

    These are disparate metrics and it is hard to get a qualitative feel for a campaign’s “management quality”. I can see chernoff faces being useful here.

    Here is another example that shows chernoff faces for Quality of life


    Chad: The above link also shows how you can use colors with chernoff faces effectively.