In my opinion as a beginner statistician, most of the students have this question in mind, because they do not aware of the functions of statistics in research (as I did have before). I think we should first show the aims of a research and then useful statistical method. More importantly, most of the students in natural science are poor at mathematics because it is nonsense for them and because of unknown functionality.

Statistics and mathematics are languages rather than sciences. So we do not learn it if we do not know in which language to speak of the results of a research.

The key problem is that statistics is not appropriately included in the study subjects. It is separated and tought seperately. In some courses/practicals the students will have to calculate averages, possibly standard errors, rarely they will do a regression, and that's it. And ever there nobody explains why they have to do it (this way) ("because it is just done like this and you will get the full points when you do it like this").

To my opinion, the teachers of specific subjects (e.g. physiology, neurology, ecology, whatsoever) have to be better trained in statistics and they have to be able to explain the scientific reasoning behing empirical sciences. Additionally, I would wish a better real interdisciplinary work between statisticians and other subjects. Why not analyse the data generated in practicals in the statistics class, where also some thorough background could be provided? So my answer is neither to teach statistics before or after research but rather along/parallel with research, already starting in the study curriculum where own data is generated in practicals (what is currently been treated as an orphan).

Yes, please: basic and advanced courses on statistics/ data analysis for scientists, as compulsory subjects and not as electives (at least in the first years when students are clueless about what is important), with examples from the particular field of study (which can hardly be provided by a math faculty, let alone if there is no cooperation with the respective science institute).

Contemporaneous to the specific sort of science and its various contents it always is advisable to teach students more generally what science and research are all about; what data are and what you can do about them; how to spot, prove, test relationships between phenomena and which means are necessary to do that. But such concepts are quite often not early and extensively enough addressed in study programs (e.g. in the geosciences, biosciences).

I have found in teaching practice ,students jump to Methods before doing the approach to the problem or problem definition. While the approach is towards new knowledge statistics is a quantitative tool for validation. One cannot substitute one for the other.

@Dr Mazumdar: I suppose you can't address aspects of problem definition and ways of approaching a problem in science without introducing concepts of data analysis and statistics as well. In fact the latter may give you a whole new idea to look at problems and even help you to sort out which problems should be approached in the first place (e.g. because their solution is better feasible than those of others or not as costly in terms of effort).

There is a huge misunderstanding plaguing the appreciation of statistics in natural science: many scholars think of statistics as the cream used to decorate cakes, thus in order to be decorated a cake must be obviously been already cooked and thus they say 'wow now I understand what statistics (cream) was for', if this is the view point clearly you ask for 'practical application of statistics using actual research content of a particular field'. The point is that statistics IS NOT THE CREAM to be put over the cake but is the indication of the relative quantity of the ingredients in the cake's recipe.

In other words, statistics is nothing more nothing less than the quantitative version of the kind of reasoning any scientist (no matter what his/her specialty is) applies to the problem at hand. In this sense it comes both before and together (you are totally right Jochem) the learning f the subject: the more I learn about a specific subject the more I can adopt useful and effective metaphors taken by the day-by-day life to describe the case I'm studying so I can describe my investigation in general terms like 'If the molecule X is inducing a differentiation in this cell population I should observe the cells becoming more and more similar among them as for the parameters Y1, Y2, Y3, this corresponds to a decrease in variability that I could study by means of coefficient of variation (SD, entropy measures..).' Now the statistics learning allows the scholar to link the 'mental image' of the 'committment' toward a differentiated state with the 'organizing' of cell population along a 'specific common profile' and thus with a suitable measure of relative homogeneity..now the same kind of thinking can be applied to marketing (creation of markets'niches), to the loss of biodiversity caused by an order parameter like a very salty terrain allowing only few vegetal species to survive or anything else. Statistics is a matter of 'educated representation style' that needs both the knowledge of what I'm observing and the representation technique, otherwise the risk is the attitude of many scholars of a blind use of a ready-made algorithm applied on a purely consuetudinary base.

Mike, I am biased to qualitative !

Theory building happens through this and only then can you 'test' theory ? If there is no theory how can you form a hypothesis ? In fact a mixed method is always appreciated since a qualitative approach may be required to answer some of the grey areas in the quantitative analysis .

In my opinion, Qualitative and Quantitative methods are both necessary when conducting a research. Some research will choose Qualitative method. Others may use Quantitative one or mixed-method. It depends on the research design! However, Statistics is very powerful tool for generalizing the conceptual framework for instance.

From a logical point of view, statistics needs to be taught before students turn up in labs. You simply need to have a basic understanding of methods and theory beforehand: statistics before analyses, molecular biology before PCR, physiology before the practical and so on.

As a student, I experienced like so many other a dry statistics course with rather synthetic data. Much better would have been faculty researchers "donating" some actual data for real world problems that have been published.

Nowadays, I'm quite involved in statistics, and I experience that people - students and researches alike - don't understand the necessity and "beauty" of statistics. After all, only statistics tells you if your data actually mean something. This is also a problem with the researchers; quite often they don't have a profound understanding of statistics, hence are unable to teach their students the logic behind the selection of specific tests and the basic background of the formulas involved.