Our research process
At Talamo, our mission to support individuals with Special Educational Needs (SEN) is powered by a commitment to creating a statistically valid, reliable, and accurate dyslexia testing and cognitive profiling tool. This tool is designed using best-in-class methods to be trusted by candidates, schools, and parents and we've followed a 5 stage process when creating and improving the test:
Our stages of research
Creating the test material
We start by collaborating with globally renowned publishers and psychometrics experts to develop our test content. This stage aims to replicate as much of the depth of a full diagnostic assessment as possible in a more concise format. Through this collaborative process, we crafted a diverse test battery that accurately measures areas such as verbal, visual, and non-verbal reasoning, phonological processing, working memory, processing speed, reading, and spelling. We purposely created more content than we’d need to choose the most effective content for each area.
User testing the material
This phase involves thorough user experience tests to assess each test's clarity, appropriateness, and difficulty. We split our user testing into stages to make improvements between each session. Feedback gathered during this phase informs the refinement or removal of test content ahead of more extensive data collection, ensuring that the foundation for our assessments is solid and user-friendly.
Creating the model
After completing Stage 3, we had a large battery of tests which showed good statistical significance in differentiating between dyslexic and non-dyslexic. The next stage was to build a model which would use multiple tests to accurately spot dyslexia.
We experimented with multiple statistical methods but settled on a Logistic Regression as the most effective. With this model in place, we then used Recursive Feature Elimination to identify the tests which were most effective at spotting dyslexia (also taking into account how long the test is). We built multiple models based on this approach and then chose the final test battery based on the model's sensitivity and specificity score. Our current model now sits at 95% accurate.
As we build our data set, this number will only increase.
Considerations
Care when collecting data
Our commitment to creating a diverse and representative dataset shines through our collaborative efforts with schools across the UK, ensuring optimal testing conditions and maximum engagement. We ensured a demographically accurate spread and attended each school numerous times to oversee the collection. See the schools we've partnered with.
What next?
Our dedication to refinement and innovation is ongoing. With each test, piece of feedback, and new piece of research, we continually enhance our tool's reliability and effectiveness.
At Talamo, we're committed to supporting neurodiversity through continuous innovation, research, and an in-depth understanding of the communities we serve.
