Our own earlier systems to simulate data for statistical exercises were based on standard models, such as using a linear model to generate outcomes based on parameters to which random Normal variability was added. In contrast the Island relies heavily on differential equations to capture changes in physiology at a more basic level and then links these with various statistical models as needed. While the population simulation moves in monthly steps, the experimental simulation updates at 30-second intervals. At this level each student has their own copy of the Island, tied to their login, so that changes they make to Islanders through experimental treatments are independent of changes made by other students. Similarly, such changes made by the students do not affect the underlying historical simulation. For example, a student can never kill an Islander through their actions since this would mean that they would then need a separate timeline in the historical simulation.This gives students freedom in the experiments they design. Some of the tasks apply treatments, such as giving the subject a tablet containing 1 mg of alprazolam or making them swim freestyle for 200 m. These do not show any output in the interface but the tasks have changed the state of the Islander in the simulation. Since the simulation happens in real time the students need to wait to observe effects. Other tasks then produce data, such as measuring blood pressure or pulse rate, or taking a blood or urine sample to detect a particular substance. The onus is on the student to develop the protocol for applying the treatments and making the measurements. In addition to tasks that mostly behaved like measurements,indoor growers the students could also design a survey for their Islanders to complete.
Questions that students were interested in asking were added to the system, along with a range of standard survey questions that we typically ask the students themselves. The Islanders would take longer to complete longer surveys, discouraging students from just asking all the questions. Some Islanders were predisposed to lie on the surveys, particular for questions related to age and weight. Students can always tell they are lying because the actual age of each Islander is displayed on the profile and there is a measurement task for weighing an Islander. There is also a chance that an Islander will decline to respond to a survey. This chance varies between the different cultural groups on the Island making a pattern in non-response bias for students to explore.The instructor can see all tasks assigned by an individual student to each Islander they used as well as Islanders that have been visited by the student but who have not had any tasks assigned to them. These monitoring tools provide a useful level of accountability for individual student work and also give the instructor a broad insight into how the environment is being used by their class. The Islanders have a pair of chromosomes that they inherit in the usual way from their parents. The 256 ‘genes’ on each these chromosomes are used to determine a variety of attributes, including the physical characteristics seen in Figures 1, measures of disease susceptibility, and other parameters required by the task simulations described in Section 3.2. We give students access to this genetic information through a task that generates the analog of a micro-array image for one or both of the chromosomes. Students can use this task to carry out studies that mirror micro-array techniques in the real world. For example, Bulmer and Meiring describe a student project that looked for evidence of a gene linked to diabetes on the Island. In that study the student obtained micro-array images for 10 Islanders with diabetes and 10 without, giving the results shown in Figure 3. The question is whether there is any systematic difference between the intensity levels expressed in each set of images. Figure 4 shows a more quantitative summary of the results with side-by-side box plots for each of the 256 genes appearing in the micro-arrays.
For convenience we label the positions with B000 at top left, along to B015 at top right and then continuing by rows to B255 at bottom right. The student carried out a two-sample t test to compare the levels between the subjects with diabetes and those without for each of these. The four strongest effects were for genes B116 , B041 , B186 and B118. Figure 5 shows the box plots for these four comparisons in more detail. For B116 the intensity distribution for non-diabetics seems uniform while it appears systematically lower for diabetics. We now use this student study as an example in class. Of course the p-value for B116 needs to be treated by caution since it arose from a large number of multiple comparisons. In our introductory course we use the very conservative Bonferroni adjustment to the p-values, whereby B116 becomes non-significant, but the overall example illustrates to students the practical issues involved in this kind of screening as well as an area of current research in the discipline of statistics itself. A great advantage of this approach is that new students can replicate the study if they want, or search for other similar genes. Student feedback to the Island has been very positive. In our context the role of it has been to replace real experiments and this is reflected in many comments such as that “they were interesting and a great way to find results of experiments. It made the experimentation process nice and easy to conduct”. However students were also engaged with the Islanders beyond a basic tool for generating data: “I liked how we were able to see their whole history on their profiles; it was interesting seeing some of their troubled past”. As mentioned earlier, we do have a tension between reality and fantasy in our design and it was interesting to read comments on the reality aspect, such as that “the Islanders were a little too real, especially as they improved reaction times after repeating the action. We really had to think of them as real people – which I suppose was the whole point”. In contrast there have not been open comments on the ‘unreal’ aspects, such as the Islanders with elven ears or the unusual disease names. We suspect that students are used to these features in computer games and are not surprised by them. This is an interesting area for future studies. The consistent negative feedback has been on the Islanders sleeping each night. While some students were interested in studying sleep,vertical hydroponic system as in the dextroamphetamine example above, for the majority of students the fact that the Islanders go into an uninterruptible sleep each night is often a nuisance. At this stage we are continuing to include this constraint, as part of our general philosophy summarized in the following section.
However we have made the Islanders go to bed a bit later each night and have introduced some tasks that can be performed on sleeping Islanders, such as the various blood tests and a simple polysomnography tool, so that students are not completely stuck if they have left their project until the night before it is due. In general it is not surprising that the Island has been successful in engaging students with the task for which it was designed. We are interested now in evaluations from other users of the Island in contexts different from our own. For example, Linden et al.give outcomes of a research grant that has investigated the use of the Island as a tool in teaching clinical trial design and management. Edwards and Crowther give an evaluation of the Island in a health systems management course where the focus was not on statistical reasoning at all. Such projects give insight into the transferability of the tool while also feeding back ideas to further expand the models included in the simulated environment. Bulmer gives more details regarding this collaborative approach. Immuno staining of the slices was performed as described previously. Briefly, after washing in PB , sections were cryoprotected in 10% sucrose and in 30% sucrose in PB for 15 min and overnight, respectively, then freeze thawed four times in an aluminium-foil boat over liquid nitrogen to enhance penetration of the antibodies without destroying the ultra structure. Residual sucrose was washed from the tissue in PB and then endogenous peroxidase activity was blocked for 10 min by treatment with 1% H2O2 dissolved in PB. Subsequently, all washing steps and antibody dilutions were carried out in Tris-buffered saline. After extensive washing in TBS , sections were first blocked with 5% normal goat serum for 45 min, washed in PB for 15 min, and then incubated with the primary antibody of interest for 48 h. The following polyclonal, affinity-purified primary antibodies were used in the present study: rabbit anti-DGL-α ; rabbit anti-MGL ; and rabbit anti-MGL. The specificity of the DGL-α-INT antibody was confirmed by the lack of immunostaining in hippo campal sections derived from DGL-α knockout mice. The specificity of the two MGL antibodies was supported by immunostaining in HEK293 cells transiently expressing a V5 epitope-tagged MGL; by the lack of immunostaining in neurons preincubated with 5 µg/ml of the corresponding immunizing protein; and by the identical staining pattern in hippo campal sections at the electron microscopic level with the two antibodies raised against independent epitopes of the MGL protein. After primary antibody incubation, human and mouse hippo campal sections were washed extensively in TBS , then treated first with biotinylated goat anti-rabbit IgG for 2 h, washed again three times in TBS, and then incubated with avidin-biotinylated horseradish peroxidase complex for 1.5 h. This step was followed again by washing in TBS and in Tris buffer , and finally the immunoperoxidase reaction was developed using 3,3- diaminobenzidine as chromogen and 0.01% H2O2 dissolved in TB. After the development of immunostaining, sections were washed in PB, treated with 1% OsO4 in 0.1 M PB for 20 min, dehydrated in ascending series of ethanol and acetonitrile, and embedded in Durcupan. During dehydration, sections were also treated with 1% uranyl acetate in 70% ethanol for 20 min. After overnight incubation in Durcupan, the sections were mounted onto glass slides and coverslips were sealed by polymerization of Durcupan at 56 °C for 48 h. Light microscopic analysis of immunostaining was carried out with a Nikon Eclipse 80i upright microscope , and light micrographs were taken with a DS-Fi1 digital camera. For electron microscopic investigations, selected immunoreactive profiles and regions were photographed and re-embedded for further ultrathin sectioning. Series of consecutive ultrathin sections were collected on Formvar-coated single-slot grids and counter stained with lead citrate for 2 min. Electron micrographs were taken at 30,000– 50,000× magnifications with a Hitachi 7100 electron microscope. For immunofluorescence double staining, after freeze-thawing and intense washing, the sections were first blocked with 5% normal donkey serum for 45 min, and then incubated with mouse anti-NeuN and either with rabbit antiDGL-α , or rabbit anti-MGL , or rabbit anti-MGL primary antibodies for 48 h. Afterward, the sections were washed again in TBS three times for 15 min each, then incubated with secondary antibodies Alexa 594- conjugated goat anti-mouse IgG and DyLight 488- conjugated donkey antirabbit IgG for 2 h. Excess secondary antibody was washed out three times in TBS, and three times in 0.1 M PB for 15 min each. Finally, the sections were mounted in Vectashield onto glass slides, and the cover slips were sealed with nail polish. Image acquisition was performed with an inverted Nikon Eclipse Ti-E microscope equipped with an A1R confocal system. Images of double labeling were obtained of a single focal plane by a 4× objective in sequential scanning mode using a four channel PMT detector. For the adjustment of digital light and electron micrographs, Adobe Photoshop CS2 was used. In all imaging processes, adjustments were done in the whole frame and no part of an image was modified separately in any way.