There are several major challenges that you face as soon as you try to get the average biology undergraduate to engage with statistics. Many are nervous of anything even vaguely mathematical, have no background in the scientific method or probability, and may even struggle with computers. As a result, traditional approaches to teaching them statistics, usually involving a walk through the long bestiary of statistical tests complete with their mathematical underpinnings and detailed assumptions has often left them cold and uninformed. Practicals and tutorials spent wrestling with the mechanics of generating statistics often result in students that can reliably produce the required number, but have focused so much on the process they have no idea why they have done it or what it might tell them. Furthermore, often the details of making sure that your study is robust in the first place are only briefly touched on. My increasingly strong view is that a possible solution to the problem is, rather than trying to get students to engage in everything at once through a dedicated statistics course, it is better to break the problem down, dealing with one component at a time. Despite what my statistics teacher apparently believed, it is perfectly possible to appreciate what an F ratio and its associated P value are telling you (and are not telling you), without being able to partition sums of squares by hand.......So why do we so often insist that the latter is a necessary prelude to the former? Far easier to engage students with the mechanics once they understand the point.
My current view is that the problem can be broken down into 4 steps. At each step, the aim is to teach concepts, and improve student confidence such that the next step is seens as valuable.
Step 1: Basic concepts
To teach students the basics of good design, with a focus on asking clear questions and thinking about what data they might collect to answer them. Concepts of replication and pseudo-replication, and observation vs experimental evidence. Encouraging the students to use visual methods to examine their data, as a prelude to statistical tests.
Step 2: From picture to test.
To show students how a statistical test can provide more objective information that simply looking at the picture. To show that the appropriate statistic becomes obvious if you have a clear question and carefully collected data, To introduce the concept of P values, and explain why it is useful.
Step 3. Introduction to the mechanics.
To teach students how to do their own simple tests focusing on a small toolkit. To get them confident designing studies that they could address with the limited toolkit, and to explicitly link question to design to statistic.
Step 4. Extending the toolkit
Introduce the broader range of statistical tests, model fitting, other forms of inference, advanced model criticism.
In particular, I think that getting students confident with the interpretation of tests is not only far more important that a rigorous training in the tests themselves, but also aids in that aspect of statistical training when student inevitably do need to do some statistics for themselves. Similarly, providing training on a limited tool kit, and then challenging students to design studies within these constraints is a powerful way to ensure they think explicitly about how they will analyse their planned study (and may act against the common view that with so many possible tests at our disposal, we will find something that works).
I am incorporating this structure into my own teaching. Steps 1 and 2 are currently embedded in labs for early level biology courses on our Zoology program. This is intentional, to make clear that these approaches are a core part of everything they do, not an add on that has to be studied separately.