The police Department (in my country) usually analyses traffic accident data according to factors that contribute to the occurrence of such accidents. (See Table below, where for example the factors are: i. poor mechanical state of vehicle (MC); ii. driving after taking alcoholic drinks (DD); iii. Overloaded vehicle (OL); and Careless driving (CD), for instance, driving while talking on a mobile phone. Apart from providing counts (or percentages) of occurrences of accidents per month and the accompanying factors (‘causes’), the traffic data appear to say less than they probably should. For example, in January the OL counts are 4 and the MC counts are 24, but the effects of OL and MC are not obvious from the data. What other methods of analysis are appropriate for such data? Would principal component analysis (PCA) shed more light?
Table: Counts of traffic accidents (not real)
==============================================
Month MC DD OL CD ...
=============================================
Jan. 24 10 4 34 ...
Feb. 31 15 5 32 ...
March 40 34 12 45 ...
April 15 23 7 23 ...
… … … … … ...
=============================================
If your data type like that, may be you can analyze via time series analysis in order to PCA. You can read first about economic research method and statistic literature, specialy about regression.
I mean, PCA is pretty much always "appropriate" in that the assumptions pretty much always hold. But does it actually give you any interesting information here? I'm not certain that's the case. Usually you use it when you have thousands of variables and want to pull out vectors which act together, or the vector of variables which cause the most change. For a small system like this? You can easily run each possible regression (vs time, i.e. a time series analysis) and get similar information.
Thanks Chris. The main question is what method works best for such data? PCA appears to be a distraction there. Also the Table does not provide all the data.
Analysis of accidents could be a science by itself. PCA could be a starting point for a first analysis of the data.
I think that the dataset is very small because there are external factors that have to be considered, for example weather (rain), visibility, type of road, etc. because at this moment, in this table, you only consider vehicles or human factors.
CD usually have to be the main factor but in this table there is not so clear.
Hi George,
Road crashes are almost always multi-factorial. That is, given a good crash reporting system, you would almost certainly find that the table you present is far too simple. A crash might result from mechanical failure and intoxication and some factor not accounted for such as road condition or weather. The last has already been mentioned. To properly analyze such data, you should also know the baseline prevalence of such things as overloading, poor vehicle condition, etc. For instance in Australia (where I work) almost all vehicles are subject to regular inspection, so it is rare to see one with worn out tires, smoking exhaust and missing parts. Therefore a mechanical failure would be a less common occurrence here. However, in your country it seems that vehicles in poor condition are common, and so mechanical failure is the second most common factor in crashes. It is probably impossible for you to change the crash reporting system, but gathering the information about the prevalence of risk factors in the population and comparing that with the observed crash rate might allow you to state that while many vehicles on the road are in poor condition, only a few are overloaded. This means that the "exposure" of the average driver to poor mechanical condition is much greater than to overloading, and that the latter may be a more important factor as it is less common in the population. If you can obtain information about multiple factors in a crash, perhaps you should look at "multiple cause" analysis.
Jim Lemon
Transport and Road Safety Research
University of NSW
Hi Jim
Many thanks for sharing. I like your proposal to gather information about the prevalence of risk factors in the population and to compare that with the observed crash rate.
Hi Mohammed
What is the aim of the research? Two reasons: Contribution to the community and the second motivation is research.
First, my Department (Department of Mathematics) is hoping to put together short training programs that assist the police department collect, analyze and report its statistical data more accurately (at least less superficially). That, we hope will create more awareness and enable motorists to improve the state of their vehicles.
Second, some of our students want to do their projects in this area of transport and road safety. At the moment, we don't yet have a clear method for analyzing the data .
Dept. of Mathematics
I work as a Policeman in Mijas PD (Spain). I can send you our report model for accidents, but it is in Spanish, in a few days.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis