Employers can now record pretty much everything that we do at work.
These days, they can even find out about what we get up to when we’re not at work too.
This is because HR has ‘big data’ – otherwise known as large, complex data sets that require very fast databases to process them. But the data itself isn’t enough. Instead what professionals really need are the big insights that such data can help to provide.
While advocates of HR analytics claim that we’re making progress in this regard, the examples that they provide of existing implementations in action tend to focus on descriptive metrics that are rooted narrowly within a single HR department rather than across a number of HR functions.
So it would appear that being routinely able to use psychological insights and statistics to tackle important HR issues is still some way off. But what would the signs that we were making progress look like?
For one thing, it would be possible to link data from relevant HR sources together in formal causal models to describe how workers could transform internal inputs into goods and services for consumption outside of the organisation.
To do this, however, would require HR professionals to act like scientists. They would also need to employ a three-step approach based on informed speculation, data gathering and statistical analysis to test their original suppositions:
1. Informed speculation
At this stage, HR practitioners would need to clearly specify their hypotheses about the relationships (or lack of relationships) between different aspects of the workforce that could be of interest to predict or improve upon. Such hypotheses might include the relationship between reward systems and turnover, for instance.
To develop plausible models of cause and effect that could subsequently be tested, however, it would be necessary for professionals to draw on insights based on their own experience rather than rely on the services of a statistical wizard.
If they were lucky enough to have access to academic literature as well as general practitioner material, it is likely that their models would be all the better informed for it too. They would also be less likely to rediscover facts that were already well-known.
For clarity’s sake, however, it is also important to express hypotheses in the form of diagrams. For example, if you draw arrows between causes and their consequences, it can help you to structure and challenge your own thoughts. It will also make the job of presenting your ideas to others in the business easier.
Also bear in mind that, although inspirational guesses gleaned when working alone might work in some instances, the process generally works better if there is broad involvement across the business.
People who understand business matters well can help in providing a rationale for your hypotheses. Without this rationale, an observed relationship between two data sets, even if statistically significant, is likely to be spurious.
In order to test the usefulness of your new model, however, ask yourself this: what would you do differently if the right data were available to support it? Then make a list of actions that you would take.
2. Data gathering
The second step is to gather the necessary employee data in order to test your hypothesis. Social media sites such as LinkedIn and Facebook should also be included in an employee’s ‘data trace’, which can be defined as broadly as required in order to undertake adequate testing of your informed hunches.
A close audit of the electronic data traces that most organisations gather on a typical worker usually reveals that not everything pertinent is typically collected. As a result, it may be necessary to become more systematic about what information is harvested and why.
While the HR department might already be recording a lot of information, reporting on it tends to be descriptive in nature. Such reports also tend to focus on an HR specialism such as recruitment or remuneration or particular employee attributes such as patterns of absenteeism.
Moreover, there hasn’t usually been a requirement to make the data available in a format that a statistical analyst could use in order to test the diagrams described in the previous step.
However, it is crucial that the different kinds of information collected in order to validate (or disprove) these hypotheses are arranged in such a way as to permit statistical modelling of what are likely to be very complex relationships.
Think of it as if you were representing your employees in the form of rows in a data file and showing cause and effect as columns. While your database analysts may tell you that this approach isn’t the most efficient way to store data, it tends to be the most effective format for analysing it.
3. Statistical analysis
The third step involves using statistical analysis to support or disprove your informed hunches. Such analysis is likely to comprise a blend of conventional statistical modelling, and where appropriate, various approaches taken from predictive analytics and machine learning disciplines.
The outcomes that you are trying to explain or predict such as performance, and the level at which you are making the predictions (individual, team, organisation-wide) will vary based on the HR problem at hand.
One of the benefits of ensuring that formal statistical models fit your data is that you can test whether or not your model, and the data that you use to test this model, are well aligned. In other words, you can accept or reject your model based on how it fits with the data.
This approach brings a degree of objectivity and rigour to the process in order to help you make scientifically-informed decisions about your workforce. A statistical expert will be required to help here, but be careful not to let their wizardry win sway over your common sense.
Ideally you should also ensure that subject matter experts are involved. If you find that your model is supported, the next phase is to implement the necessary change and track the effects of your intervention over time.
If your model was not supported, however, at least you should have gained some insights into what a better model might look like.
The idea here is to give HR pros some insights into how they can take big data and turn it into ‘big insights’. Although many practitioners feel that they’re doing this already, there is still a lot of work to be done to truly gain a system-level understanding of how employees contribute to business competitiveness.
But such understanding is within reach and taking a ‘practitioner as scientist’ approach to it will help you get there.
Dr Nigel Guenole is programme director for leadership and talent management at Goldsmith’s Institute of Management Studies, University of London.
