WEBVTT

00:00:00.000 --> 00:00:06.000
<b><h4>How do you measure/assess significance? Do you ever drop or ignore some data/findings?</h4></b>

00:00:06.000 --> 00:00:06.020
<b>Caroline Morais, PhD Researcher, University of Liverpool</b>

00:00:06.020 --> 00:00:39.290
So to assess the significance in my data, it's a very important point because I am using data from major accidents and they are fairly here, where situations. So some points for some problems, some root causes in their accident reports, we have lots of important descriptions from the regulators. But for other points, other problems and causes, we have very few. 

00:00:39.290 --> 00:01:12.450
So it's very difficult to assess the significance of this few data. We call it sparse data. And our institute are all explained with sparse data, not only with big data. So again, I am using some methodologies to assess this problem. And about the findings, I will never drop any kind of point. We try to use all of-- we have to use all of them.
<hr />

00:01:12.450 --> 00:01:18.190
<b><h4>Do you consider different ways of interpreting your findings? </h4></b>

00:01:18.190 --> 00:01:53.170
<b>Caroline Morais:</b> I have to consider different ways to interpret my findings because at the very beginning when I started to do verification and validation to other people's results, my results, my findings were behaving a little bit weird. And I started to change the model and trying to see how it goes, how it impacted my results. And I could see that I started to speak of this problem, to open this problem in conferences in my first two conferences. 

00:01:53.170 --> 00:02:13.610
And I could see that everyone in the community, in my community, that is human reliability analysts they were also having the same problems. But they were not exactly discussing that on the papers because everyone thought that was such a big problem that it was a problem of their research, and not a problem of the data itself. 

00:02:13.610 --> 00:02:54.574
So when we are tracking human errors, we started to see that we cannot see as it was an engineering system. I cannot compare my data and my findings with my colleagues' data and findings because they are analysing various engineering systems-based findings, and I'm looking for human errors, probabilities. So we started to see that data interpretation had to be somehow different. We had to involve some psychologists and some philosophers together with engineering. 
<hr />

00:02:54.574 --> 00:02:58.840
<b><h4>What do you do to try to avoid bias?</h4></b>

00:02:58.840 --> 00:03:26.120
<b>Caroline Morais:</b> Usually in my research topic, that is human reliability, when there are some holes in the grades, when there is some holes with data, when we have sparse data, people use expert judgment. So they ask some experts to give some judgments of what that point should be. And that's not considered very credible in my area of research. 

00:03:26.120 --> 00:04:01.440
So we started to use what we call imprecise probability. So we could quite relax and be honest that at that point, in that interval, we don't have any points, and that's going to be explicit in the result. So the result's not anymore a point, it's an interval. So in this sense, we believe that we are being transparent with society, that when they see that interval they know that we still need more data, we don't have all the data. 