Next from book: Data Science with Julia of Paul D. McNicholas and Peter A. Tait, 2019, CRC Press.
The 21st century will probably be the century of the data revolution. Our numerical world is creating masses of data every day and the volume of generated data is increasing more and more (the number of produced numerical data is doubling every two years according to the most recent estimates). In such a context, data science is nowadays an unavoidable field for anyone interested in exploiting data. People may be interested in either understanding a phenomenon or in predicting the future behavior of this phenomenon.
To this end, it is important to have significant knowledge of both the rationale (the theory) behind data science techniques and their practical use on real-world data. Indeed, data science is a mix of data, statistical/machine learning methods and software. Software is actually the link between data and data science techniques. It allows the practitioner to load the data and apply techniques on it for analysis. It is therefore important to master at least one of the data science languages.
The choice of the software language(s) mainly depends on your background and the expected level of analysis. R and Python are probably the two most popular languages for data science. On the one hand, R has been made by statisticians... mostly for statisticians! It is, however, an excellent tool for data science since the most recent statistical learning techniques are provided on the R platform (named CRAN). Using R is probably the best way to be directly connected to current research in statistics and data science through the packages provided by researchers. Python is, on the other hand, an actual computer science language (with all appropriate formal aspects) for which some advanced libraries for data science exist. In this context, the Julia language has the great advantage to permit users to interact with both R and Python (but also C, Fortran, etc), within a software language designed for efficient and parallel numerical computing while keeping a high level of human readability.
...
What is data science? It is an interesting question and one without a widely accepted answer. Herein, we take a broad view that data science encompasses all work related to data. While this includes data analysis, it also takes in a host of other topics such as data cleaning, data curation, data ethics, research data management, etc...
The place of statistics within the pantheon of data sicence is a topic on which much has been written. While statistics is certainly a very important part of data science, statistics should not be taken as synonymous with data science. Much has been written about the relationship between data science and statistics. On the one extreme, some might view data science -and data analysis, in particular- as a retrogression of statistics; yet, on the other extreme, some may argue that data science is a manifestation of what statistics was always meant to be. In reality, it is probably an error to try to compare statistics and data science as if they were alternatives. Herein, we consider that statistics plays a crucial role in data analysis, or data analytics, which in turn is a crucial part of the data science mosaic.
Contrasting data analysis and mathematical statistics, Hayashi (1998) writes: "... mathematical statistics have been prone to be removed from reality. On the other hand, the method of data analysis has developed in the fields disregarded by mathematical statistics and has given useful results to solve complicated problems based on mathematico-statistical methods (which are not always based on statistical inference but rather descriptive)"
The views expressed by Hayashi (1998) are not altogether different from more recent observations that, insofar as analysis is concerned, data science tends to focus on prediction, while statistics has focused on modelling and inference. That is not to say that prediction is not a part of inference but rather that prediction is a part, and not the goal, of inference. We shall return to this theme, i.e., inference versus prediction, several times within this monograph.
Breiman (2001b) writes incisively about two cultures in statistical modelling, and this work is wonderfully summarized in the first few lines of its abstract: "There are two cultures in the use of statistical modeling to reach conclusions from data. One assumes that the data are generated by a given stochastic data model. The other uses algorithmic models and treats the data mechanism as unknown. The statistical community has been committed to the almost exclusive use of data models. This commitment has led to irrelevant theory, questionable conclusions, and has kept statisticians from working on a large range of interesting current problems."
The viewpoint articulated here leans toward a view of data analysis as, at least partly, arising out of one culture in statistical modelling.
In a very interesting contribution, Cleveland (2001) outlines a blueprint for a university department, with knock-on implications for curricula. Interestingly, he casts data science as an "altered filed" -- based on statistics being the base, i.e., unaltered, field. One fundamental alteration concerns the role of computing: "One outcome of the plan is that computer science joins mathematics as an area of competency for the field of data science. This enlarges the intellectual foundations. It implies partnerships with computer scientists just as there are now partnerships with mathematicians".
Writing now, as we are 17 years later, it is certainly true that computing has become far more important to the field of statistics and is central to data science. Cleveland (2001) also presents two contrasting views of data science: "A very limited view of data science is that it is practiced by statisticians. The wide view is that data science is practiced by statisticians and subject matter analysts alike, blurring exactly who is and who is not a statistician".
Certainly, the wider view is much closer to what has been observed in the intervening years. However, there are those who can claim to be data scientists but may consider themselves neither statisticians nor subject matter experts, e.g., computer scientists or librarians and other data curators. It is noteworthy that there is a growing body of work on how to introduce data science into curricula in statistics and other disciplines (see, e.g., Hardin et al., 2015).
One fascinating feature of data science is the extent to which work in the area has penetrated into the popular conscience, and media, in a way that statistics has not. For example, Press (2013) gives a brief history of data science, running from Tukey (1962) to Davenport and Patil (2012) -- the title of the latter declares data scientist the "sexiest job of the 21st century"! At the start of this timeline is the prescient paper by Tukey (1962) who, amongst many other points, outlines how his view of his own work moved away from that of a statistician: "For a long time I have thought I was a statistician, interested in inferences from the particular to the general. ...All in all, I have come to feel that my central interest is in data analysis, which I take to include, among other things: procedures for analyzing data, techniques for interpreting the results of such procedures, ways of planning the gathering of data to make it's analysis easier, more precise or more accurate, and all the machinery and results of (mathematical) statistics which apply to analyzing data."
The wide range of views on data science, data analytics and statistics thus far reviewed should serve to convince the reader that there are differences of opinion about the relationship between these disciplines. While some might argue that data science, in some sense, is statistics, there seems to be a general consensus that the two are not synonymous. Despite the modern views expounded by Tukey (1962) and others, we think it is fair to say that much research work within the field of statistics remains mathematically focused. While it may seem bizarre to some readers, there are still statistics researchers who place more value in an ability to derive the mth moment of some obscure distribution that in an ability to actually analyze real data. This is not to denigrate mathematical statistics or to downplay the crucial role it plays within the field of statistics; rather, to emphasize that there are some who value ability in mathematical statistics far more than competence in data analysis. Of course, there are others who regard an ability to analyze data as a sine qua non for anyone who would refer to themselves as a statistician. While the proportion of people holding the latter view may be growing, the rate of growth seems insufficient to suggest that we will shortly arrive at a point where a statistician can automatically be assumed capable of analyzing data.
This latter point may hel to explain why the terms data science, data scientist and data analyst are important. The former describes a field of study devoted to data, while the latter two describe people who are capable of working with data. While it is true that there are many statisticians who may consider themselves data analysts, it is also true that there are many data analysts who are not statisticians.
Along with rapidly increasing interest in data science has come the popularization of the term big data. Similar to the term data science, big data has no universally understood meaning. Puts et al. (2015) and others have described big data in terms of words that begin with the letter V: volume, variety, and velocity. Collectively, these can be thought of as the threes Vs that define big data; however, other V words have been proposed as a part of such a definition, e.g., veracity, and alternative definitions have also been proposed. Furthermore, the precise meaning of these V words is unclear. For instance, volume can be taken as referring to the overall quantity of data or the number of dimensions (i.e., variables) in the dataset. Variety can be taken to mean that data come from different sources or that the variables are of different types (such as interval, nominal, ordinal, binned, text, etc.). The precise meaning of velocity is perhaps less ambiguous in that it is usually taken to mean that data come in a stream. The word veracity, when included, is taken as indicative of the extent to which the data are reliable, trust worthy, or accurate. Interestingly, within such three (or more) Vs definitions, it is unclear how many Vs must be present for data to be considered big data.
The buzz attached to the term big data has perhaps led to some attempts to re-brand well-established data as somehow big. For instance, very large databases have existed for many years but, in some instances, there has been a push to refer to what might otherwise be called administrative data as big data. Interestingly, Puts et al. (2015) draw a clear distinction between a big data and administrative data: "Having gained such experience editing large administrative data sets, we felt ready to process Big Data. However, we soon found out we were unprepared for the task".
Of course, the precise meaning of the term big data is less important than knowing how to tackle big data and other data types. Further to this point, we think it is a mistake to put big data on a pedestal and hail it as the challenging data. In reality there are many challenging datasets that do not fit within a definition of big data, e.g., situations where there is a very little data are notoriously difficult. The view that data science is essentially the study of big data has also been expounded and, in the interest of completeness, deserves mention here. It is also important to clarify that we reject this view out of hand and consider big data, whatever it may be, as just one of the challenges faced in data analysis or, more broadly, in data science. ...