Originally posted in 2019
The author of this post is an Organisation and People Data Analyst at GSK, and a speaker from a previous meetup. To watch @Andrea Schirru give a talk on Survivor Analyses, please click here. To be kept up to date HR and People analytics meetups around the world, please sign up to the ThinkTank newsletter by clicking here or keep an eye on our Meetup Listings Page here.
Predicting Healthcare Outcomes Using People Data Science with Professor Charlwood
In the first presentation of the evening, @Andy Charlwood (Professor of HRM at Leeds University and HR Analytics ThinkTank Academic Lead) took us though a data science example within the healthcare industry. There is currently a global shortage of nurses at the moment, but there are massive challenges to invest in the industry being a low margin business.
To develop a casual model, he considered outcome indicators related to quality, people resourcing and contextual variables. Regression analyses allowed him to view changes over time within individual homes as well as within home differences but was not very informative to take decisions. Random forest provided more immediate insight than traditional regression analysis approach. Creating a partial dependent plot in Python he could see more clearly that having a more experienced and skilled workforce will likely improve quality in nurses homes.
However, the magnitude of the quality improvements is unlikely to make this a cost effective intervention. Professor Charlwood suggested that if we wanted to learn more from observational data, we would need to use it as a basis for experimental and quasi-experimental analysis.
On 18 June, @Andy Charlwood and Dr Danat Valizade will be hosting a free webinar, talking attendees through the use of Data Science in HR, highlighting the techniques used in their talk. To register please click here.
Seagulls and Bombs: Predicting Performance in the Workplace with Andreas Kyprianou
In his presentation, @Andreas Kyprianou, Head of Talent, shared how at Bank of America they optimise assessment selection to predict if someone is going to be a good employee. Starting from Schmidt & Hunter’s key concept (1998) of predictive validity of assessment methods, that throughout more than 30 years proved that only GMA (general mental ability) & structured interviews work in practice, he focused on task performance (what) and contextual performance (how). He used ROC (Receiving Operating Characteristics) to predict whether someone is going to be a high performer taking all the test scores (the higher on top of the curve the better the prediction power of the test) and then created a binary variable (>50 high performers, all others). This way he proved that 66% of the time the numerical test would be right (‘what’) and, going a step further, tested the interrelationships between the tests through a Chi-Square and Regression Analysis. He found out that if the 3 variables numerical testing, AC-grouping exercise & 1st stage interview are all present the likelihood of predicting high performers is 75%, if one of 3 is missing it drastically falls down. There is a high degree of interrelations between those variables.
However, with this approach, it was not possible to predict well the contextual performance. Therefore, as a result, for the task performance he requested improved versions of numerical testing, group exercise and 1st stage interview. He focused on training interviews and group assessors to identify impression management techniques and implemented a cut-off point of 80% at numerical testing to improve sifting prior to assessment centre. Within the contextual performance, because it was not predicted by anything, he introduced situation judgement testing and are now looking at different things, e.g. behavioural assessments, gamification tools and personality assessment testing.
To watch the video (for free) please click here.
To join us a the London meetup, or to find HR and People Analytics meetups near you around the world, please sign up to the ThinkTank newsletter by clicking here or keep an eye on our Meetup Listings Page here.