Preregistration of analyses

Preregister at least two analyses that you vow to present in your final report, no matter the results (significant or not). Do not perform these analyses, check they are significant, and then “preregister” them. That defeats the purpose. You can learn a lot from associations that are not statistically significant!

Keep in mind that for your final report, you must provide at least one model showing patterns or relationships between variables that addresses your research question.

Write your preregistered analyses in the pre-registration.qmd file in your project repo. Make sure to include any questions you have for your project mentor.

Example preregistration

Hypothesis 1

Chocolate-based flavors are more popular on the East Coast than in the Midwest.

Analysis: Run a linear regression where we input region (as a dummy variable) and output proportion of chocolate-based flavors (# chocolate-based scoops / # total scoops) sold on an average day. The Midwest dummy will be our reference variable, so we will test whether \(\beta_{\text{East coast}} > 0\).

Hypothesis 2

The cost of ice cream increases with distance from Ithaca, NY.

Analysis: For data where each row represents a different location, we run a linear regression where we input the distance of the location from Ithaca, NY, and output price (in dollars) of ice cream. Because our distance measurement will have signed values (e.g. anything West of Ithaca would be negative miles and East would be positive miles), we will test whether \(\beta_{\text{Distance}} \neq 0\).

Evaluation criteria

Category	Less developed projects	Typical projects	More developed projects
Preregistration statement	The preregistered analyses could be performed using the data collected, but it is not clear how they fit in the context of the real-world application from which the data originated.	The preregistered analyses are contextualized by the real-world application to a certain degree. The analyses are not described in a way that persuades the reader their results would be interesting, whether or not they turn out to be statistically significant.	The preregistered analyses reflect deep and critical thinking about the real-world application from which the data originates. The analyses are described in a way that persuades the reader that their results will be interesting, whether or not they turn out to be statistically significant.