4
$\begingroup$

Would the knowledge of the following topics suffice for a crash course on fundamentals of applied statistics?

  • Basics of Probability
  • Descriptive Statistics
  • Estimation Theory
  • Hypothesis Testing

The targeted audience are those with little or no exposure to statistics. The course should bring about a clear understanding of what is statistics and its basic outline. It should also serve as a start for those who are interested in the subject but don't know where to begin.

The "applied" part comes with the implementation of the above theories in practical examples and exercises.

$\endgroup$
  • $\begingroup$ You have to teach said course? $\endgroup$ – Repmat Jan 4 '16 at 7:20
  • $\begingroup$ Yes, @Repmat, but the syllabus is on my own accord. $\endgroup$ – Ébe Isaac Jan 4 '16 at 7:27
  • 3
    $\begingroup$ Who's judging what "suffices" and on what criteria? I'd say that no one is well prepared for applied statistics without knowing more about data production and data management, but there's only so much you can put in one course without strain or complaint. The premise of a "crash course" is arguable itself: applied statistics is a long slow art. $\endgroup$ – Nick Cox Jan 4 '16 at 10:40
  • 2
    $\begingroup$ Hard to know for sure what you intend to cover under each of those headings, but your list reads like an introduction to theoretical Statistics - as a pre-requisite for intermediate applied Statistics perhaps. $\endgroup$ – Scortchi Jan 4 '16 at 15:28
  • 1
    $\begingroup$ I am concerned about the potential for this thread to devolve into a list of personal opinions, but at the same time I am optimistic that this community can contribute some well-reasoned informed replies. Instead of closing the thread, I have provisionally made it CW (in the same spirit as other threads that have no unique best answer but can be useful to readers). Ébe Isaac, some elaboration in the post of what you mean by "applied statistics" would be welcome. As @Scortchi has suggested, most of your topics don't look applied at all. $\endgroup$ – whuber Jan 4 '16 at 16:13
4
$\begingroup$

If these are subject-matter-experts I would spend more time discussing:

  • How to ask the right question in a way that it can be answered.

  • How to collect data so that you can answer the question.

  • How to select a statistical procedure that can answer your question.

  • What the result of the statistical analysis means and how to know if it is of any practical value.

  • When to call in professional statisticians and how to effectively work with them.

$\endgroup$
1
$\begingroup$

The targeted audience are those with little or no exposure to statistics.

My favourite audience!

It should also serve as a start for those who are interested in the subject but don't know where to begin.

Your choice of textbook is going to be crucial. My strategy is to have one very short stats textbook (e.g. Urdan, Stats in Plain English), and to do everything else through code and tutorials.

The "applied" part comes with the implementation of the above theories in practical examples and exercises.

This is the crucial part. Make the examples relevant: no synthetic data, and of the right dimensions for your audience (don't teach flower or airplane data to would-be survey analysts).

  • Basics of Probability
  • Descriptive Statistics
  • Estimation Theory
  • Hypothesis Testing

That's one way to put it. I would stress:

  1. Data. What it is (observations, variables, samples), how is it produced.
    • Data collection, sample design, measurement issues.
    • Question(s) -> Data -> Question(s) -> Data -> etc.
  2. Distributions. How do we describe one variable, several variables.
    • Your descriptive stats are here.
    • Tons of EDA (exploratory data analysis) / data visualization always helps.
    • Concentration indices, quantiles, ECDFs. Invaluable.
  3. Estimation. When we have $\bar x$, how do we estimate $\mu$?
    • A bit of probability and estimation theory here.
  4. Inference. When we observe an association, how do we measure its robustness?
    • Hypothesis testing happens here. More probability here.
    • Downplay $p$-values, emphasize standard errors.
    • If you have time, models.

My strategy:

Anticipate that students will forget (3) and (4). That's alright, the next course will take it up again. But (1) and (2), your students need to be familiar with the jargon, and should be critical quantifiers, who know that data are dirty, that sampling is complex, that "raw data" is an oxymoron, etc.

General philosophy:

Source.

$\endgroup$
  • 1
    $\begingroup$ How I hate this "tidy" buzz word after taking a data science course. I'd never use it in a professional setting $\endgroup$ – Aksakal Jan 6 '17 at 1:55
  • $\begingroup$ Is "tidy" just not in the workflow, or the jargon is different? (If I understand correctly, in my area this would be something like "QC and standardize", the latter w.r.t. format) $\endgroup$ – GeoMatt22 Jan 6 '17 at 2:11
  • $\begingroup$ "Tidy" means easily manipulable in R (i.e. rectangular). You can forget about it if your course/area uses something else. $\endgroup$ – Fr. Jan 6 '17 at 2:14
  • $\begingroup$ There is no reason the "tidy" principles would not be useful in other contexts than R. $\endgroup$ – kjetil b halvorsen Jun 23 at 21:00
  • $\begingroup$ "Tidy" turns out to be exactly the layout of data I've regarded as standard for just about 50 years.... $\endgroup$ – Nick Cox Oct 7 at 12:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.