Version 1.5.1

• backward compatible with Matlab 2015b

Version 1.5

• You can now do parameter estimation with exponential discount functions.
• Added validation of behavioural data upon import, to help avoid errors.
• You can now specify what figure export formats you prefer.
• Should work on PC's now, and have updated install instructions on the wiki.
• Added help on the wiki to explain the optional arguments to the core model parameter estimation procedure.

Version 1.4

Quite a large update here in terms of amount of behind the scenes effort. A number releases will follow soon which finalise a few in-development features.

Meaningful changes

• Run analyses using 1 line of code. Note that any existing analysis scripts you have will need to be updated. Sorry, but this is going to be a better, simpler, and more future-proof way of doing it.
• Inference using STAN. Working, but will be polished in the next release.
• Added unit tests. Increases confidence in code and makes it easier to extend the toolbox without unknowinly breaking stuff.
• Many small bugs are fixed. Things should run a bit smoothly now, but please do submit an issue if there you find a bug.
• Increased code commenting, where appropriate.
• Installation should be easier and more robust, and instructions on the wiki are updated.

Behind the scenes

• Many code simplifications and robustifications.
• Simplified models through modelling group-level inferences as an additional (unobserved) participant. Avoids a lot of duplication. (see #89)
• Data now internally represented in long-form. This makes it easier to write STAN models as it doesn't support ragged arrays (ie when participants have different number of trials).
• Better application of some of the SOLID principles of OOP

Version 1.3

Posterior predictive checks

• We calculate the proportion of responses that the model can accurately predict.
• We get a distribution over this proportion predicted score.
• We can use this as a quantitative exclusion criterion.

Version 1.2.2

New models

• Semi-hierarchical model for estimating log discount rates. This 'mixed' model is recommended when you are estimating discount rates for a heterogeneous population, for example if your dataset consists of multiple groups or conditions. It does however still use hierarchical estimation for error params (alpha, epsilon) as these are unlikely to vary systematically between groups. If you think that is not the case, then you should use the existing 'separate' models which have no hierarchical inference at all. More info about this model will be in the wiki.
• Gaussian Random Walk model. This model is highly experimental and treats the discount function (indifference points) in a non-parametric fashion, as a Gaussian random walk as a function of delay. This model is only suitable for experimental methods which have attempted to identify the indifference points at a variety of delays, i.e. multiple questions at each delay (DB) values.

Other changes

• I've updated priors over log discount rate logk based upon examining quite a lot of data from my lab.
• Continued minor bug fixed and attempts at making the code clean.

Version 1.2.1

New features

• Exported point estimates are now JASP-friendly. Now it's easier to collect data > run your analysis in this toolbox > export point estimates > analyse in JAPS or SPSS etc.
• Choose your favourite point estimate type. Choose from posterior mean, median, or mode. You'll have to justify this decision in a paper of course.
• New plots. Even more plotty goodness to understand your data.

Behind the scenes updates

• Toolbox dependencies are now auto-downloaded/updated. You may need to install git once, then forget about it.
• Even more code clean-up and refactoring. Further along the quest for clean code.

Version 1.2

Key changes

• New model to estimate discount rates. This is very handy when you don't want to estimate the magnitude effect. Now you can estimate the discount rates, log(k), assuming the 1-parameter hyperbolic discount function.
• New non-hierarchical models I've also added non-hierarchical versions of the magnitude effect, and the discount rate models. This means that inferences about discount rates are conducted independently for each participant. This is handy when you know in advance that your participants do not come from a homogenous group. For example when there are different participant populations, or different experimental conditions.
• Behind the scenes code refactoring. This should make the code more readable and it should be much easier for you to develop your own models. And easier for me to allow inference using STAN in a future release.

Version 1.1

Key changes

• Helpful instructions. See the wiki for lots of information such as how to vet participant data and checking the reliability of the inferences.
• Introductory video. See the video here to get an idea of how quick and easy it is to use the software.
• Point estimates are now posterior mean. Previous estimates were based on the posterior mode, but this has now been changed to the posterior mean.
• Additional plots. A couple of additional plots are now produced to get a greater insight into the data.
• A lot of behind the scenes code refactoring. This is ongoing, but is is making the code more modular and expandable. This will help future goals of adding additional models and adding the ability to run inferences with STAN.