Upon graduation I was amazed to learn that what I thought data analysis meant and what most businesses thought it meant were largely different things. For the majority of business, data analysis consists of looking back at what has happened in a period of time, segmenting the data, using averages or other basic statistical metrics to make a graph. For me, data analysis is about using the information collected and leveraging predictive models like Ordinary Least Squares (OLS) or Logistic Regression (LOGIT) to understand how each variable affects the outcome independent of the influence of other variables. Even a step beyond that, it is possible to understand how the same amount of change in a variable might not always affect the outcome the same amount.
For example, if Cup of Noodles doubled in price, say from $0.20 to $0.40, would you consume dramatically less of it? Likely not — although that change represents a 100% increase in price, it likely doesn’t affect your consumption all that much. However if the price increased to $1.00 or $2.00, you might consume dramatically less. This same principle applies to the salaries offered candidates or the price of we pay for clicks on job boards. Increasing the salary offer to a candidate by $5,000 could increase the probability that the offer will be accepted by someone making $40,000, but it may not increase the probability for someone making $400,000.
By understanding and predicting the non-linearity of how individuals or firms react to changes in the market, we can better allocate resources and make more informed decisions. In the context of recruitment advertising, by comparing the probability of a user clicking on varying job titles for the same job, we can better understand what is appealing to job seekers in a purely scientific fashion and more effectively utilize budget. By taking our focus away from hindsight and focusing on insight and foresight, we can stretch our budgets further and increase the effectiveness of our efforts. Below is an example of a non-linear probability model that in this example would predict the probability of a given viewer clicking on a job. We cannot use an OLS model in this context because the outcome is perfectly binary (i.e. They either click or they do not, no in-between).
(This graph does not represent actual research and is purely for the sake of demonstration.)
If the X-axis represents the CPC of a job posting and the Y-Axis represents the probability that a viewer will click on it, we can see that as we increase the CPC, the probability of a click also increases. However, it is also important to note that the increase in probability of a click is dramatic between .50 and .60 but between .65 and .70 the change is much smaller. Also we can see that the probability of a click could be different between two job titles even at the same CPC, represented by the red vs. blue lines.
What are your data-driven strategies when it comes to recruitment advertising analytics? Share with us in the comments!