We have talked about ‘out-takes’ but what about ‘outcomes’?  We got the coverage, we changed people’s attitudes but did they then go to the website, call our call centre, walk into the shop, sign up for the trial or buy that iPhone?  There are a number of approaches that can help with this.  First and most simply we can plot a media analysis metric (say favourable coverage) against our outcome (say website visits).  We may feel this shows a correlation, but can we prove it, can we quantify this?

Luckily, if you have Excel, there are a couple of easy functions that can help.  CORREL() and PEARSON() both effectively do the same thing which is to give you something called the correlation coefficient.  The correlation coefficient is a number that goes from -1 to +1 that shows the strength of a correlation between two sets of data.  +1 is a perfect positive correlation (as one thing gets bigger the other thing gets bigger), -1 is a perfect negative correlation (as one thing gets bigger the other gets smaller) with zero being no correlation.

The correlation coefficient while showing a strength of correlation, doesn’t show how statistically significant it is.  The significance is a function of the correlation and the number of data points (which is why more data points generally gives better results).  There are a couple of advanced functions in Excel will provide this information, but it is also available from statistical textbooks and websites.

There is a danger with using this type of correlation work to analyse data over time.  The maths behind the analysis assumes that two points in time are totally independent of one another when in reality this is not true.  In the graph below we have an advertising campaign where the sales always spike two weeks after the adverts.  The correlation in this case would be close to zero.  By shifting the advertising forward by two weeks the correlation becomes close to one.

Advertisers have got around this problem by introducing the concept of ‘ad-stock’, where the effectiveness of an advert declines with a set ‘half-life’ over the following weeks.  In the example below the advertising has a half-life of two weeks.

Correlation work on its own can work in simple situations where PR is the major factor affecting an outcome.  Unfortunately in many real-world situations, things are more complicated where there are many factors at play from multiple marketing activities through to effects from pricing or seasonality.  To this we need to look at econometrics, the subject of the next post.

Next: Part 6 – Econometrics

Previous: Part 4 – Market Research