Using tablet-based testing to reduce cheating in large-scale student assessments in Andhra Pradesh

2019-02-01

2019-03-15

Our core objective is to compare cheating in schools which were tested using paper-based assessments to those tested using tablet-based assessments.

We will measure this in two ways: (a) through direct comparison with the retest audit in 120 schools and (b) through an analysis of suspicious response patterns in the item-level data from the full assessment, as done in the testing literature elsewhere.

We set up a large-scale experiment covering all schools with at least 5 students enrolled in Grade 4 in one district in February 2019.

Randomization was carried out not at the level of the individual school but at the level of an academic cluster which typically covers multiple villages. This was to keep the mode of testing unchanged within a single educational market so that schools within the same community could be fairly compared in student report cards. Out of 284 clusters, 196 were assigned for tablet-based testing and the remainder 88 to paper-based testing. All schools within a cluster were assigned to be tested using the same protocol (whether paper or tablets).
There was one pre-determined deviation from the initially-assigned status. Academic clusters, for the most part, nest villages/urban wards but not always. In such cases, if cluster A and B were in different treatment arms, the treatment status of all schools in the village was reassigned to be the same. For instance, if a village had 6 schools, 4 of which are in cluster A and 2 in cluster B, the treatment status of all schools in the village was reassigned to be the same as Cluster A. Such reassignment affected 191 schools out of 2462 schools, which are disproportionately urban areas. In the final sample, 768 schools were assigned for paper-based testing and 1694 to tablet-based testing.

Students, across both treatment arms, were assessed in three subjects – Mathematics, Telugu (the official state language), and English – using the same test papers in February 2019. The tests were designed centrally and intended to capture a wide range of variation. All questions in the test booklets were multiple choice items. In each subject, to deter student cheating, three test booklets (“sets”) were created which had 80% of items in common but with some distinct items and with a distinct ordering of test questions. In schools assigned to paper-based assessment, the test booklets were sent with clear instructions for schools to administer the tests and for the answer scripts to be returned to the Mandal Education Office (the administrative unit above schools). Tablets were taken to the school by a Cluster Resource Person (CRP), each student was given an individual tablet to work on and the test booklet for each subject was decided by the software directly.

The analysis in this study will compare the responses of students who were assessed using tablet to those who were assessed using paper booklets. A potential problem in this comparison is that students may underperform in tablet-based assessments, not only due to potential distortion in paper-based tests but also because they are not familiar with tablet-based tests. Thus, we randomly sampled 120 schools, spread equally across the paper and tablet-based testing arms, and retested students using a traditional paper assessment but with external proctoring and test administration by the research team. This retest was conducted within two weeks of the original assessment but could only be completed in 117 schools.

On a computer, done by Principal Investigator

At teh academic cluster level

Yes

284 clusters

In the final sample, 768 schools were assigned for paper-based testing and 1694 to tablet-based testing.

196 clusters in the tablet-based testing arm, 88 clusters in the paper-based testing arm.