Wikipedia:Reference desk/Archives/Science/2019 March 11

Welcome to the Wikipedia Science Reference Desk Archives
The page you are currently viewing is an archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

March 11

Remnants after an aviation crash

One could expect that after high-speed aircraft impacts with ground (including CFIT), like recent Ethiopian Airlines crash, most if not all remnants at the side would be charred or burned due to instant explosion and/or resulting fire after smashed fuel tanks. However on many photos various things (like paper instructions, personal belongings, etc) often look unscathed. Why is that? 212.180.235.46 (talk) 18:19, 11 March 2019 (UTC)

There are many factors that can affect how widely spread the flames are (if any). Some are:

• How full the fuel tanks are.

• Where the fuel tanks are located. Wing tanks, for example, will tend to burn the fuselage less, especially the parts far from the wings, like the cockpit and tail.

• Where the engines are located. A tail engine would require a fuel line and thus increase odds of fire in the tail section.

• The terrain and weather. Trees may cause the wings to be torn off. Smooth ground would be more likely than rocks and trees to leave the plane intact. The moisture level in the terrain and weather conditions would also affect whether a fire would spread over the ground. In hilly terrain, spilled fuel could be expected to flow downhill, either before or while burning.

• The angle at which the aircraft strikes the ground. A shallow angle (perhaps with bounces) and a high speed may result in the plane being torn into sections, as in Uruguayan Air Force Flight 571. In such a case, if there was a fire, it wouldn't be likely to affect all sections and all the debris that fell out during the crash. A plane crashing vertically is more likely to be totally incinerated, but such a crash isn't likely to be classified as "controlled flight into terrain", more likely to be "suicide by pilot", terrorism, or some other cause.

• Whether there are items on board likely to aid a fire, like (punctured) oxygen tanks, flammable fabrics, etc. SinisterLefty (talk) 21:50, 11 March 2019 (UTC)

Press coverage of airline disasters also includes those which explode at high altitude and these obviously produce a wide debris field, often with many minimally-damaged items. Perhaps you have seen "many photos" with these kind of results. Or is your interest exclusively in "CFIT-type" accidents? Crashes towards the end of a flight obviously will involve less fuel. Crashes just after take-off often involve more intense fires where most of the airframe and contents are totally lost. Martinevans123 (talk) 22:55, 11 March 2019 (UTC)

Outcome prediction of events when there aren't many previous cases to build on

When you got lots of previous and similar cases (say team A plays against team B for the N time) you could apply statistical techniques to predict outcome with a certain margin of error (also statistically assessed).

What about science dealing with stuff when you don't have loads of previous cases? What techniques do we have to reasonably deal with future outcomes? --Doroletho (talk) 20:01, 11 March 2019 (UTC)

That is an incredibly wide field and I think it depends on specific details. If you have a significant data set of basically the same type of experiment (such as very generally "team vs. team"), then methods exist to make predictions for specific circumstances that have never arisen. Very generally multivariate statistics can be used to create a model that explains observed correlations between variables, and allows for the prediction of possible data points not present in the actual data set. In a sort of overly simplified version of this, a sports fan could simply create a table of which teams both A and B have played against, who won those games, and make a guess from there. Basically, "Team A always loses to X, but B always beats X, so B is better than A." There are obvious pitfalls to that type of model. In other circumstances, you can try to make an inference by generalizing the problem. We often do this in such things as guessing the chance that an experimental drug will reach FDA approval. When that drug has only just passed its first phase I trial and has not been properly assessed for effectiveness yet, it may seem impossible to make a prior assumption. However, we can generalize, asking "of all the drugs (of this type | for this illness | other generalization) that passed phase I trials (with similar results), how many ultimately reach FDA approval?" Really, this is basically assuming a multivariate model, where the features of a drug or its development are predictive of its success, even where this variable combination has never been seen before. With any of this, you would always be more confident in interpolations rather than extrapolations. That is, when the individual variables being considered for the prediction are within the range of variables for the experiments you have data for. Someguy1221 (talk) 01:09, 12 March 2019 (UTC)

I'm not so sure that predicting the probability of a particular outcome is the same thing as "predicting the outcome". ←Baseball Bugs ^{What's up, Doc?} carrots→ 07:39, 12 March 2019 (UTC)

Let me phrase that a bit differently: "It may be possible to predict the specific outcomes, but not by using statistics." For example, when mixing two chemicals which have never been mixed before, it's possible to predict the outcome. Acid-base reactions lists one type of reaction which could be predicted to occur if one of the chemicals is an acid and the other is a base. In the case of the chemicals, as long as we can assume no contamination and known concentrations, the result should always be the same. This is not true in sporting events, with a classic example being the World Series in baseball, where the same two teams can play repeatedly, with each game having a different outcome. There are just too many unknown variables, like how sore a given pitcher's arm is, to accurately predict the outcome.

Examining how car insurance rates are calculated for new drivers may also be helpful. They look at characteristics like sex, age, whether they completed driver's training, and location (which covers the local weather, road conditions, population density, theft rates, etc.). So, here they do uses statistics, but statistics for those general characteristics, rather than for the specific driver. SinisterLefty (talk) 12:10, 13 March 2019 (UTC)

Weibull distribution and Bayesian Statistics. The first keeps a running score as you go, and builds the statistical model at the same time. Engineers use it. The second is deeply weird and interesting, to this engineer. Greglocock (talk) 08:08, 12 March 2019 (UTC)

The Drake equation is an example of analysis of the identifiable but as-yet unquantified factors needed to predict the probability of a case that has, as yet, never happened, namely discoveries of active, communicative extraterrestrial civilizations. DroneB (talk) 09:24, 12 March 2019 (UTC)