Package to compute the ProjectBased Assessment estimates of student and rubric proficiency.
Project description
Project Based Assessment
The project based assessment library allows the practitioner to estimate difficulty and ability parameters when using data from rubric rows.
The library contains the following methods:
 getResults
 DisplayResults
 SaveResults
getResults and DisplayResults take the following parameters:
 dataset: A pandas Dataset containing the columns "k", "student", "rubric", "bound". The "k" column is the rubric level the given student reached on the given rubric row. The "student" column is a student identifier. The "rubric" column is a rubric row identifier. The "bound" column is maximum "k" value possible on the given rubric row.
 c: A float between 0 and 0.5 indicating the portion of the bootstrapped EDF to extract. For instance, specifying 0.025 would produce the 95% confidence interval. Default is 0.025.
 rubric: A bool flag indicating to treat the rubric rows as blocks instead of the unique students in the bootstrap. Defaults to False.
 n: The number of iterations in the bootstrap. Defaults to 1000.
 linear: Uses a simple linear combination of the rubric and student items instead of a sigmoid function when set to true. Defaults to False.
 columns: A list of column names to include in the model. The column names cannot be in common with any of the rubric row identifiers. Defaults to None.
 no_students: If True, the model will not include student (s_i) estimates. Defaults to False.
 no_questions: If True, the model will not include question/rubric row (q_j) estimates. Defaults to False.
SaveResults includes the same parameters as getResults and DisplayResults but has three additional parameters:
 rubricFile: File name/path for the rubric results. Defaults to 'rubric.csv'.
 studentFile: File name/path for the student results. Defaults to 'student.csv'.
 outputFile: File name/path for the summary output results. Defaults to 'output.csv'.
All methods return the following:
 Rubric difficulty estimates as a pandas dataframe. Additional interpretation columns are provided in this dataframe that will be described below.
 Student ability estimates as a pandas dataframe. Additional interpretation columns are provided in this dataframe that will be described below.
 Bootstrap confidence intervals and PValues as a pandas dataframe. PValues are only provided when estimating the nonlinear model as they will always be zero for the linear model (by construction the estimates are constrained between 0 and 1 in the linear model).
 The number of times the bootstrap routine could not find a solution (if any).
 Number of observations.
 Number of parameters.
 Akaike information criterion
 Bayesian information criterion
 McFadden pseudoR^2
 Likelihood Ratio test statistic
 ChiSquared PValue of the model (i.e. Wilks' theorem)
 Log Likelihood value
getResults only return these values as a tuple. DisplayResults returns the values as a tuple and prints the results to screen. SaveResults returns the values as a tuple, displays the results and saves the results to CSV files.
The rubric difficulty and student ability pandas dataframes return estimates along with columns used for interpretation. The following columns are provided:
 AME k=i: The average marginal effect of k=i. This is provided for all possible bins (i between 0 and the highest bin). This procedure calculates the marginal effect for a given estimate conditioned on k=i for all observations impacted by the estimate. The average is then calculated. These values will sum to zero.
 ACP k=i: While average marginal effect is the standard approach to interpreting MLE results (especially in a logit or probit context), we don't think they are particularly useful in this model. Therefore, the application also provides columns for the average conditional probability of k=i. Given the subset of the data used to calculate AME, this is the average probability of k=i given the estimated value. When the dataset is balanced (all students have a score for all rubric rows), these values will sum to 1. Note that the top bin is capturing the censoring effect. Therefore, it is common that a substantial probability is estimated for this bin.
 Average Logistic: This estimate is only provided when estimating the nonlinear model. It is the average of the probability function given the estimated value. It uses the same subset of the data used to calculate AME and ACP above. In terms of interpretation, it is the average probability of failure to proceed to the next bin. Therefore, it will equal ACP k=0.
 Average Marginal Logistic: This estimate is only provided when estimating the nonlinear model. It is the average of the marginal probability function (derivative) given the estimated value. It uses the same subset of the data used to calculate AME and ACP above. In terms of interpretation, it is the change in the average probability of failure to proceed to the next bin.
 Average Discrete Marginal Logistic: This estimate is only provided when estimating the nonlinear model. It is the average of the discrete marginal probability function given the estimated value. It uses the same subset of the data used to calculate AME and ACP above. In terms of interpretation, it is the change in the average probability of failure to proceed to the next bin. This is very similar to the Average Marginal Logistic above and will often produce very similar values. However, this value is calculated by calculating the probability with the variable in question and without the variable in question and then taking the difference. This is a common approach when calculating the discrete marginal effect in logit and probit models. The Average Marginal Logistic is calculated by taking the derivative. Where these values tend to diverge is when the estimate is at an extreme. Consider a student who has extremely low ability such that they fail every trial. In this case, the derivative will be near zero as it is the instantaneous slope at the extreme value. However, the Discrete Marginal Logistic will be very large as the average change in probability over the range is large.
Background and Use
This package is based on the estimator presented in "Assessing Proxies of Knowledge and Difficulty with RubricBased Instruments." There is a video demonstrating using this package in Google Colab and a video explaining the paper.
Installation
You can install the package from either PyPI or the Conda repository:
pip install ProjectAssessment
or
conda install c tazzben projectassessment
Project details
Release history Release notifications  RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for projectassessment0.3.3py3noneany.whl
Algorithm  Hash digest  

SHA256  4d27882b0ff21369efd7d1eb26a4941898cc23be3afbcc9f1b708fb78898167c 

MD5  77f53afd7c75e76e953c61ecbbeb8ad3 

BLAKE2b256  772a2b86780d52b26870fa0467148ff31864cec4451fa59183a23c8f88c08b80 