I have spoken to a number of recruiters and headhunters, and they agree: LinkedIn recommendation system is not useful to them because it’s highly biased.
The fact that references are not anonymous means that the writer cannot be honest, simply because the writer could be risking to jeopardize a whole relationship for no benefit whatsoever to him/herself. There is research suggesting that even the smallest of criticisms can jeopardize a relationship.
Who has not read this: “John recommends Peter”. “Peter recommends John”. The act of reciprocity is pure human nature. But it makes recommendations less valuable, simply because the writer is more willing to reciprocate.
The user is able to choose whether to show a reference or not. That creates a selection in two ways. Firstly, the user is only going to show the references that will be shown. Secondly, new users will not even get to write a bad reference (or even average) because they know that it will not be accepted by the user, so it creates a significant selection bias.
4. Unstructured Text
Text takes a long time to read and understand, especially when people often write those long references. It’s hard to know what’s important and because the writer is not especifically requested to write about something in particular, like “weaknesses” or “behaviours at work” or “ethics”, etc. most people end up providing a generalistic and positive view on the person
5. Qualitative benchmark
When we say that someone is perseverant, we probably mean that she is more perseverant than others. The benchmark is important. On LinkedIn references there is no benchmark, it’s just an absolute opinion, and purely qualitative, with no numeric table or scale that readers can refer to.
The problem with this biases is that, effectively, they cannot be used for recruitment or to check people’s reputation or ability to work well. They become an additional data point which contains very limited value.
At TraitPerception we tried to fix many of these biases. Firstly, being numerical where the benchmark is everyone else in the system. Secondly, being anonymous and statistically significant, which also reduces reciprocity because you don’t know who wrote something good or bad about you. Also, everyone in your network can provide an opinion and the user is not allowed to decide whether to show it or not.
Our solution is not perfect, and we are always trying to find better ways to measure people’s reputation and provided fairer scales, but we believe that our world, where we do more and more transactions with people we have not met before, needs these kind of solutions so we can trust people and assess them more accurately and fairly.
Juan holds an MBA from Chicago Booth with high honors and an MSc in Electrical Engineering from UPM.
His hobbies are capoeira, a brazilian martial art, drawing and tennis.
“Reward excellent failures”