Financial incentives or app-based support: improving academic progress through a communication intervention

Second semester students from Norwegian higher education institutions receive an email from SELF with either
1) a gain-framed message highlighting the incentives in the student financing system for credit completion,
2) a loss-framed version of the same information or
3) an invitation to participate in a study success program delivered through a web application.
The control group is not contacted. While treatments 1) and 2) are delivered in April, treatment 3 starts in January.

Revision June 2026:
Due to low take up of the Study Success program, we decided to revise the treatment protocol for treatment 3 and add a fourth treatment arm.
Revised 3) In addition to the invitation to participate in the Study Sucess program sent in January, students also received the loss email (T2) together with a link information video from the Study Sucess program in April.

4) A loss-framed message highlighting the incentives in the student financing system combined with two information video from the Study Success program.

The e-mails of the revised treatment 3 and the new treatment 4 are delivered in April parallell to treatment 1 and 2.

Control (T0): The control group does not receive any contact. The control group is only
invited to the endline survey.

Incentives gain frame (T1): The incentives gain frame receives an email in April 2026
presenting the information about the incentives for credit completion embedded into the student financing system and a reminder that it is not too late to start now.

Incentives loss frame (T2): The incentive loss frame group receives the same information as the gain frame group, except for that it is emphasized that failing to complete credits means that they lose a grant, as opposed to earning a grant if one completes the credits.

StudySuccess program (T3): This treatment group receives an invitation at the beginning of the semester to participate in the study success program. Students who enrol gain access to three modules delivered through a web application over the course of a semester and two short reinforcement videos. Participants receive text message notifications whenever a new module or video becomes available.

The web application combines several components: goal setting, realistic planning, soft-commitment (Himmler et al., 2019, Brade et al., 2024); information about study strategies (Delavande et al., 2022); and brain functioning and stress management (Yeager et al., 2022). Using an end-of-semester survey we will measures students’ attitudes towards these components, as well as their use of planning and study strategies and their perceived stress levels, in order to shed light on the mechanisms through which this composite intervention affects outcomes.

There is substantial uncertainty about the number of students that will sign-up to participate in T3 making it hard to know whether we have the statistical power to have different treatment arms within this treatment. As a result, we will soft-launch T3 in week 3, 2026 with a random sub sample of T3 participants. If sufficiently many students onboards the web application during this soft launch, we will divide T3 into two groups in the full launch in week 5. One group (T3a) will see the app with a soft-commitment, and one group (T3b) will see the same content without the soft commitment.

Revision:
After the soft-launch of T3 we understood that signup in T3 would be low. We thus decided to revise the T3 protocol, drop treatment T3b and launch an alternative treatment (T4) with one third of the initial T3 sample.

StudySuccess program + incentives loss frame + video (T3): Students are invited to the study success program as described above. In addition, we add a contact point and send the loss message combined with one video link based on the Study Success program. These are the re-inforcement videos initially inteded for program participants. The video either targets students learning strategies (T3-learning) or stress mindset (T3-stress) and is randomly assigned.

Incentives loss frame + study strategies and stress mindset videos (T4): In the new T4 students receive the loss framed incentives message together with two video links with content of the Study Success program. The content are the same as in T3-stress and T3-learning.

2026-01-19

2026-06-19

We consider the following set of outcomes at the end of the second semester:
(1) Three separate measures of academic performance:
- Administrative records on completed credits,
- Missed exams and
- grade average

(2) Two survey measures on academic stress.
- Perceived stress scale (Cohen & Williamson (1988)
- Academic stress scale (Murberg, T. A., & Bru, E. 2004, Rege et al. 2025)

Completed credits are based on the institutions reporting to the Norwegian State Educational Loan Fund. All Norwegian higher education institutions are obliged to report this and there is a uniform system for the reporting across institutions. The Norwegian State Educational Loan fund convert student loans to grants based on the number of credits completed and thus this is considered a reliable source of information. Students who do not have any credits registered are assumed to not have completed any courses that semester and will be coded as zero.

Information about exam participation and performance is obtained from the exam records of the Norwegian Directorate of Higher Education. Observations are at the exam level, but we aggregate them to individual-semester level.

Missed exams are measured by the number of exams the student was registered for, but for which the student did not meet. Students can sign off the course two to three weeks before the exam, and if they do so they will not be registered as not met.

The performance on the exam is captured by the student’s grade. Norwegian higher education use one of two grading schemes. (1)Letter grades between A and F, where A is the highest grade, E is the lowest passing grade, and F is failed. These grades are transformed to the numbers 5 (A) to 0 (F) before computing the grade average. (2) Pass/fail. This latter grade scale applied to about 19 percent of the exam observations in the pilot study. Grade average is computed as the average grade of exams with a letter grade, and exams with Pass/fail are set to missing. Passed courses are captured by completed credits.

Academic stress:
The measures for academic stress is based on Murberg and Bru (2004) and Rege et al. (2025) and the perceived stress scale (Cohen & Williamson (1988)).

Exact question wording in the survey will be uploaded as an attachment to this pre-analysis plan prior to data collection. The items within each survey measure will be aggregated by taking their unweighted averages.

Based on survey measures:
Study behaviour:
(1) Self-reported hours spent on study activities (self-study and lectures) per week
(2) Self-reported planning, commitment and preparedness (Delavande, 2022)
(3) Use of study strategies and division of self-study time across strategies (Delavande, 2022)

Mindset
(4) Stress mindset (Yeager et al, 2022)
(5) Growth mindset (Belief that ability grows with effort from Gaumer, Erickson & Nooan, 2016)

Financial incentives awareness
(6) Financial incentives as motivational factor during the semester and knowledge about the role of credits for grant acquisition) (researcher developed)

Revision:
Outcomes for study behaviour 2 and 3 were altered to fit with the new video treatments. The outcomes are now:
(2) Self-reported preparedness (single item question)
(3) Use of study strategies (self-developed)

The exact survey items will be uploaded as an attachment to this pre-analysis plan prior to data collection.

We conduct a person randomized RCT with around 25 000 first year students at Norwegian higher education institutions in the spring semester 2026. These students receive an email from the Norwegian Student Loan Fund. Depending on the treatment students receive information about the incentive structure they are facing when having a student loan or receive an invitation to participate in an online study success program, providing them with information on how to study more effectively and handle stress better. Students in the control group are not contacted. We observe administrative outcomes at the end of the semester and also collect survey outcomes on stress.

Sample:
The intervention is conducted with the entire cohort of first-year students who started a bachelor’s or integrated master’s degree program from Norwegian higher education institutions that meet the following criteria:
- Is in their second semester in spring 2026 at a Norwegian higher education institution
- Was first time registered as a higher-education student at SELF in fall 2025.
- Is younger than 30 years old
- Is a full-time student
- Does not live with their parents

Students are approached in the second semester of their study program.

Empirical strategy:
The analytical sample will consist of individuals who met our pre-specified criteria at the time of randomization. The first three criteria are constant over the observation period. The final two may, however, change. Students may, at any time during the year, decide to change their study load, by signing off courses and reporting it to the Student Loan Fund. Our point of departure is that they intended to be full-time student at the start of the semester, irrespective of their later decisions. Our pilot intervention revealed that in a sample of 5000 students, only a handful of students were not registered as full-time students in the final data.

The financial incentive in T1 and T2 are contingent on students not living with their parents. This is again a decision that may change throughout the semester. If participants have changed their living arrangements prior to the distribution of the financial incentive information they will be excluded from analysis of the treatment effect of T1 and T2.


Treatment comparisons:
We test the following main hypotheses:
H1a: Providing students with information about the incentive structure embedded in the student financing improves their academic outcomes, but may increase academic stress. (Pool T1 & T2 and compare it to the control)

H2: Providing students with a study success program targeting psychological and behavioral barriers to goal attainment improves their study outcomes. (Compare pooled T3 to control).

H3: The effect of information on study outcomes and academic stress depends on the framing (Compare T1 to T2).

H4: A soft-commitment increases the effect of the study success program om academic performance (Compare T3a to T3b).

H5: Providing students with a study success program has a larger effect on study outcomes than information about incentives (Compare pooled T3 to pooled T1&T2).

Revision treatment comparisons June 2026:
H4 is removed as treatment T3b was not launched. Two new hypotheses are added:
H6: Providing students with a loss framed message about the incentive structure embedded in the student financing combined with videos targeting psychological and behavioural barriers to goal attainment improves their study outcomes. (compare T4 to control)

H7: Combining the loss frame message with videos targeting psychological and behavioural barriers to goal attainment improves students' outcomes more than a loss framed message alone (compare T4 to T2)


Control variables:
To increase power, we will control for baseline values of credits completed. In addition, we have access to socio-demographic controls such as age, gender, marital status and citizenship (foreign or Norwegian), as well as information about the study programme and institution. In addition, we will include a measure of procrastination tendencies captured by the student’s application date (proxied by the funding decision date) to student financial support.

We will define and include control variables based on their power to predict variance in the respective outcome in the control group and show robustness checks for inclusion of all the variables.

When analysing the subsample that responded to the end-survey we will in addition have access to the following control variables: whether one parent has higher education, living situation and self-reported credits signed up for in the beginning of the semester. Students who responded to the survey, but did respond to these questions will be indicated with “missing” on the given control variable.

Heterogeneity
We will investigate heterogeneity along the following dimensions:
(1) Credits earned in the preceding semester, more specifically, whether the student obtained the nominal 30 credits in fall of 2025 or not. We hypothesise that students who already are behind nominal progression are more in need of an intervention and that the effect is stronger among the students who obtained less than 30 credits.
(2) Gender.
(3) Procrastination tendency captured by the funding decision date for study support (Brade, 2024). We hypothesise that the effect will be stronger for procrastinators.

More details about procrastination analysis:
Students may apply for financial support from the Norwegian State Educational Loan Fund from the date they receive confirmation of a study place (July, 20) until November, 15. Applications can be submitted at any point during this period and funding decisions are typically made without delay. The first payment is made august 5 and then in monthly instalments. For students who apply after the semester has started, financial support is paid on an ongoing basis following approval. Everyone who applies prior to November, 15 receives the same total amount so the application date do not have financial consequences in terms of amount received. Under the assumption that individuals with stronger procrastination tendencies are more likely to delay administrative tasks, we use the funding decision date as a proxy for procrastination. Specifically, we classify students based on whether their funding decision date falls before or after the median.

The intention is to do the heterogeneity analysis for all three treatments. However, heterogeneity analysis in T3 is dependent upon sufficient sign up.

We will account for multiple hypotheses testing using List, Shaikh, and Vayalinkal (2023).


Potential additional analysis if take-up is low:
The main analysis of T3 is an intention to treat analysis. However, the ITT is not informative if the take-up rate is low. Thus, in the event that the take-up rate is low we will pursue methods that allow us to compute an average treatment effect on the treated. Our preferred alternative is carry out a Complier Average Causal Effect (CACE) analysis, which relies on standard instrumental variable assumptions to identify the local average treatment effect on compliers. For a CACE-analysis, we use the random assignment to invitation as an instrument for program participation and invoke the standard exclusion restriction, i.e., that the invitation affects outcomes only through participation.

Conducted by the National Student educational Loan fund

student

No

Approx. 25000 students

Approx. 25000 students

Original:
T1: 4000
T2: 4000
T3: 6000+1000 in test launch
T3b: 6000

Revision:
T3-stress: 4000
T3-learning: 4000
T4: 5000 (test launch group included)

Control group (T0): 5797 (remainder of population)

We conduct power analysis using optimal design software aiming for a power of 80% and a false discovery rate (alpha) of 5%. For H1, we compare the pooled T1 & T2 (N = 8000) to the control group (N=4000). This sample gives us a MDES of 0.06 standard deviations after correcting for the imbalanced group size. For H2, we compare the pooled T3 (N = 13 000) to the control group (N=4000). This sample gives us a MDES of 0.05 standard deviations after correcting for the imbalanced group size. For H3, we compare the T1 (N = 4 000) to the control T2 (N=4000). This sample gives us a MDES of 0.064 standard deviations. For H4, we compare the T3a (N = 7 000) to the control T3b (N=6000). This sample gives us a MDES of 0.05 standard deviations. For H5, we compare the pooled T1 & T2 (N = 8000) to the pooled T3 (N=13 000). This sample gives us a MDES of 0.04 standard deviations after correcting for the imbalanced group size. For the survey outcomes, we, conservatively, expect a response rate of 10%. Adjusting the sample size accordingly, assuming equal participation accross treatments leads to the following MDES using the comparisons outlined above. For H1, we compare the pooled T1 & T2 (N = 800) to the control group (N=400). This sample gives us a MDES of 0.19 standard deviations after correcting for the imbalanced group size. For H2, we compare the pooled T3 (N = 1 300) to the control group (N=400). This sample gives us a MDES of 0.16 standard deviations after correcting for the imbalanced group size. For H3, we compare the T1 (N = 400) to the control T2 (N=400). This sample gives us a MDES of 0.20 standard deviations. For H4, we compare the T3a (N = 700) to the control T3b (N=600). This sample gives us a MDES of 0.16 standard deviations. For H5, we compare the pooled T1 & T2 (N = 800) to the pooled T3 (N=1300). This sample gives us a MDES of 0.13 standard deviations after correcting for the imbalanced group size. We expect to improve power with the addition of control variables Revision June 2026: Outcomes registry data: H4 is removed. Sample size numbers and MDE in H2 and H5 are updated to account for the lower sample size in T3. And power calculations for H6 and H7 are added. For H2, we compare T3 (N = 8 000) to the control group (N = 5797). This sample gives us a MDES of 0.05 standard deviations after correcting for the imbalanced group size. For H5, we compare the pooled T1 & T2 (N = 8000) to T3 (N = 8 000). This sample gives us a MDES of 0.05 standard deviations after correcting for the imbalanced group size. For H6, we compare T4 (N = 5 000) to the control (N = 5797). This sample gives us a MDES of 0.06 standard deviations. For H7, we compare T4 (N = 5 000) to T2 (N = 4 000). This sample gives us a MDES of 0.06 standard deviations. Survey outcomes: For H2, we compare T3 (N = 800) to the control group (N = 579). This sample gives us a MDES of 0.15 standard deviations after correcting for the imbalanced group size. For H5, we compare the pooled T1 & T2 (N = 800) to T3 (N = 900). This sample gives us a MDES of 0.14 standard deviations after correcting for the imbalanced group size. For H6, we compare T4 (N = 500) to the control (N = 579). This sample gives us a MDES of 0.17 standard deviations. For H7, we compare T4 (N = 500) to T2 (N = 400). This sample gives us a MDES of 0.19 standard deviations.