The mission of »Streuspanne« was formulated in 2018 by blog founder Sascha Feth as follows: »I would like to provide engineers in particular with intuitive access to statistics without presupposing basic mathematical knowledge. To do this, we look at a wide variety of topics from unusual angles.« A lot has happened since then. Time for an interview.
© iStockphoto
Our blog »Streuspanne« has now existed for over two years. Time for an interim balance right at the beginning of the year. In this interview, our bloggers Sascha Feth and Jochen Fiedler look back, talk about developments, goals, the motivation behind it, and the statistical future.
The interview is conducted by Esther Packullat from the Communication Team of Fraunhofer ITWM.
Sascha Feth:
Statistics doesn't exactly have the reputation of being particularly entertaining. Our customers in the automotive industry usually see it that way, too. We wanted to bring non-statisticians closer to the topics in a relaxed and entertaining way. Then Corona came along and we saw parallels to statistics in vehicles. Where we usually deal with data gaps in warranty data, now it was missing data around mild Corona histories that were not reported.
We were surprised ourselves by the great interest in the first corona posts and then decided to continue working on the topic.
Jochen Fiedler:
I had already looked at various statistics and figures relatively early on in the pandemic, especially to be able to assess what was happening. The opportunity to deal with it professionally, so to speak, and to deal with statistical evaluations came in very handy to me, which is why I am also very happy to participate in the blog.
Jochen Fiedler:
The pandemic has made the complicated interplay between statistical representation of reality and its meaning very clear to me. Ultimately, we as statisticians are only trying to represent aspects of reality; but what that means in concrete terms or what should follow from it is far from clear. As an example: There are various people who want to deduce the harmlessness of Corona by the quite small seeming statistical total mortality. But they keep ignoring hard facts, such as the fact that in Bergamo, deceased people had to be taken away by trucks by the hundreds and thousands, and that health care had collapsed. Ultimately, any statistical investigation must be measured against such realities.
Sascha Feth:
In February, I also thought it was a variant of the flu, where perhaps mortality might be a touch higher. After all, what I can see now from the numbers, I underestimated it. Likewise, it showed me once again how easy it would be to infer even gross crap from data. In some places the opposite of the truth. Unfortunately, this is also statistics: Laymen can very quickly derive false statements.
Sascha Feth:
In general, of course, these are things like the effectiveness of vaccination, whether one is immune after surviving an infection, etc. But also smaller ambiguities, such as whether there are differences in mortality between the sexes, etc. Here we just have to wait until there is more data available.
Jochen Fiedler:
We know very little about what contact patterns look like, how and when they change, and what all the reasons for that might be. In the first wave, the measures worked so well in this country partly because there was a relatively high level of fear, following the fresh images from Italy. However, things like fear cannot be measured properly, which is why it is always difficult to predict the exact effect of a measure.
We also don't know how exactly average testing behavior can change. There were very few tests over Christmas and New Year's Eve, for a variety of reasons. Without precise knowledge of these causes and their effects, however, it is difficult to make any meaningful extrapolation from the available data as to how many infections there really were.
Sascha Feth:
Unfortunately, it is predominantly catastrophic! A lot of people fall for the prevention paradox or cause and effect are confused. It is argued that everything is not so bad after all. And if the cause – i.e. Corona – is not so bad, then we don't need these measures at all – i.e. the effect. But it is precisely the measures that are the cause of Corona not being able to develop its full effect.
Jochen Fiedler:
There are a number of things that bother me a lot.. For example, the media coverage is often enough similar to a sports ticker, where the latest results and records are reported again. It would be more important to me to have a more nuanced and differentiated report on the current numerical dynamics of the pandemic and what could cause it. Over the holidays at the end of the year, the numbers were not comparable to the numbers at the beginning of December for various reasons. In reporting, precisely this and the possible reasons for it should then enjoy a significantly higher priority than announcing the latest figures, which is not the case everywhere.
Sascha Feth:
In the blog, we only work with simple models that only show that »the whole package of measures works.« The blog is supposed to be non-political and independent, which is why we don't want to go into detail about individual measures here. At the same time, we have to be careful about what »work« means. Is growth just slowed down, as it was in November, or do we need tougher measures to bring the numbers back down.
Jochen Fiedler:
As Sascha says, the blog is relatively neutral there, without concrete valuation. This is also due to the fact that it is very difficult to determine the concrete influence of measures, because a measure can only have an effect in conjunction with the willingness of the population to implement it. In other projects, we try to take into account the empirical effects of the measures and model predictions based on them.
In general, however, we can say from the experience of other countries that are halfway comparable to Germany, such as Sweden, that it is necessary to take measures to mitigate the hardships of the pandemic.
Sascha Feth:
They are very different and we try to answer all of them. Often they are questions about understanding or about the situation in other countries, because family or friends live there, for example. Unfortunately, sometimes there are also requests for changes that are deliberately intended to water down the articles. This is presumably intended to strengthen an opposing position. While we always check feedback on content, we do not allow ourselves to be instrumentalized.
Jochen Fiedler:
There were also questions about the significance of the rapid antigen tests and what one should pay particular attention to if one wants to exclude infections as reliably as possible with such tests. The statistical answer was that it depends primarily on the sensitivity of the test – the higher it is, the more certain one can be that a person tested negative does not have an infection. However, an important insight is also there that the test alone cannot represent the whole reality. If a person with symptoms has had contact with infected people, then their negative test result is much less reliable than that of a symptomless person who has had little contact with other people for weeks. Again, it is not so easy to relate purely statistical quantities to assessments of the real situation.
Sascha Feth:
A pure scientific article has only one mission: to convey information. As soberly and objectively as possible. In most cases, only experts can interpret it. A blog post, on the other hand, should make the findings accessible to more people. So it is allowed to be »sloppy« in the scientific vocabulary, as long as it serves the understanding. But it is especially important to me to say: A scientific essay can be suitable for deriving political decisions. A blog post never is. That's why we never map all the details we work out in projects on the blog. That would provide a would-be accuracy in too narrow a context.
Jochen Fiedler:
Exactly, with a blog the main focus is on presenting facts in a way that is as broadly understandable as possible. Especially in statistics, this is often not so easy, because some correlations are somewhat hidden and not immediately obvious - even as a statistician, you can get confused. As an example: No test result stands alone, because no test is always error-free. But you always have to take additional information into account and see how it affects the interpretation of the result.
Jochen Fiedler:
Especially in political or social discussions, there are always statistical arguments that are interpreted in a very flat and simplistic way. I think it's always very important to put them into a broader context, because otherwise you can get overwhelmed by the pure numbers. This often happens in discussions about crime, where this or that statistic is used to show how bad society has become. However, many aspects are often ignored, such as the development of the figures over time, the legal definition of the offense, or the social perception of the facts.
Sascha Feth:
At the moment, I prefer to choose the topics as a reaction to strange social media posts. The male pregnancy test was one such example that motivated us to create the mini-series on sensitivity and robustness. Therefore, there will continue to be many posts that are intended to put misguided discussions in the right light.
Fraunhofer Institute for Industrial Mathematics ITWM