Split in Probability Weights - Scarcity, Cognitive Load and Tunnelling

Last registered on June 29, 2022


Trial Information

General Information

Split in Probability Weights - Scarcity, Cognitive Load and Tunnelling
Initial registration date
November 11, 2021

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
November 15, 2021, 11:32 AM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated
June 29, 2022, 3:08 PM EDT

Last updated is the most recent time when changes to the trial's registration were published.



Primary Investigator

University of East Anglia

Other Primary Investigator(s)

PI Affiliation
University of East Anglia

Additional Trial Information

In development
Start date
End date
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Understanding the decision-making process is the cornerstone of better policy interventions. Broadly, poverty has focussed on systemic factors, individual factors, environmental pressures or a combination of the three. However, another side of the story remains relatively less understood  – the causal effects of poverty that change the decision-making process itself. A rapidly emerging debate between policymakers brings forth this fourth perspective. Within this framework, the shortfall of resources or poverty changes cognitive systems that ultimately affect the decision. At first glance, low pickup rates of preventative health, medications or high-interest borrowing behaviours of the poor seem actively self–sabotaging. Looking closer, they seem natural fallouts of the easily activated, challenging to suppress, interfering monetary thoughts that shape valuations and associations. Scarcity or the feeling of having less than one needs alters the decision-making process in itself. This sensitivity to 'what matters' changes preferences. Poverty triggered mechanisms make economic decisions more difficult by curtailing cognitive control.
The involuntary load presses the processing into redirecting the slower, deliberative system two towards what needs to be taken care of immediately. At the same time, other preferences get overwhelmingly guided by the faster, affective system—such recalibration in preference construction results in the rational-bias split or tunnelling. Our work is an inquiry into this dichotomy of risk preferences. We use the two-period natural harvest cycle combined with the priming of financial worries to study the decision-attribute dependent probability weighting of farmers in 800 people sample from Bwikhonge in Uganda for gains and losses in a between-within subject design. We explain the process that begins involuntarily with scarcity, has psychological implications and bifurcates the probability weighting function. For choices that resolve the scarcity at hand,  we find more expected utility consistent decisions. The probability weighting function is a straight line. In contrast, it is more curved for biased irrelevant choices.
External Link(s)

Registration Citation

Pande, Suvarna and Arjan Verschoor. 2022. " Split in Probability Weights - Scarcity, Cognitive Load and Tunnelling." AEA RCT Registry. June 29. https://doi.org/10.1257/rct.8071-2.2
Experimental Details


The design has three core elements - scarcity treatment, a cognitive load test and the risk-choice game. The experiment begins with identifying the sources of scarcity, measuring its impact on the cognitive systems and the probability weighting function for both gains and losses.
Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
We compare the differences in choice switches in the matched groups for each pair of rungs in each domain. There are three possibilities between the two attributes - -1, 0,1, and we intend to compare the number of total switches between scarcity relevant and irrelevant decisions for each probability ladder.
Primary Outcomes (explanation)
The standard common consequence ladder design choices are translated to pwf slopes by comparing the preference reversal inequalities derived from their respective value functions. Because the probability mass moved between the rung pairs remains the same, a consecutive choice switch from safe (S) to risky (R) or R to S would violate the Expected Utility theory and indicate a non-linear pwf. Given the S-R menu for any two rungs, the four possible choice switches can be coded as - S to S (SS = 0), S to R (SR = -1), R to S (RS = 1) and R to R (RR = 0). Of the four, SS and RR are EUT consistent, and the rest violate an identity (w(p) = p) pwf.

We take these basics further and compare the choice switches between the two attributes. Given the within-subject treatment at this stage, the difference would be a result of tunnelling. Hypothetically, if cognitive realignment didn't exist, the proportion of SR and RS switches would be the same between the two decision attributes. On the other hand, if scarcity relevance and irrelevance were to impact, we would expect a difference in the number of non-EUT (SR or RS) switches. Therefore, the primary outcome is the choice switch for the five rung pairs from the two attributes for each domain.

Secondary Outcomes

Secondary Outcomes (end points)
1. Effectiveness of priming - the self-reported impact of shock scarcity.

2. Cognitive load test score - time and accuracy score on a two-question task.

3. Tracking the probability weighting function for each decision attribute, given the participants' scarcity state.
Secondary Outcomes (explanation)
1. Effectiveness of priming - We introduce three real-life scenarios as our priming tool to the randomly chosen half of the sample in each session. The situation read-outs get at the individual features of financial scarcity. A Likert scale accompanies the three questions to self-report the hypothesised impact of shock scarcity.

2. Cognitive load test score - We measure the time and accuracy on a two-question task after the first intervention. The tasks are chosen to keep in mind the limitations of literacy and field constraints. Theoretically, cognitive load works through working memory, attention and inhibitory control impairment. Therefore, to quantify the causal mental impact of scarcity on decision-making systems, all our participants complete a numerical Stroop and a Digit span test.

3. Tracking the probability weighting function for scarcity relevant and irrelevant attributes separately- With the level of scarcity and decision attribute fixed, we compare the choice switches between

Experimental Design

Experimental Design
After allocation to gains or losses in both the natural periods, we randomly assign the scarcity priming treatment to half of the participants. This is followed up with a cognitive load test and a within-subject decision attribute treatment.

Experimental Design Details
Scarcity treatment - any felt scarcity can be natural, unanticipated or a combination of both. Geo-sensing data from the region shows bimodal harvest seasons. We use this knowledge of agricultural cycles as a starting point. Our prospective subjects have lived through the annual patterns year after year and have experience making decisions under these conditions. This distinct lean-plenty phase gives us a naturally occurring base for anticipated and expected scarcity. Using this annual harvest cycle for the region, we add the priming treatment to account for the unpredictability. For each of the lean-harvest base levels, we introduce real-life scenario readouts to make shock salient. The self-reported Likert scale questions differ in triggering thoughts about financial scarcity in the treatment and control groups. Each precludes one of the four characteristics of scarcity as a force (easily triggered, spontaneous, persistent and value realisation). The readouts make the unanticipated scarcity salient within each of these periods. The control group gets innocuous three scenario readouts with a similar word count. Therefore, we get a 2*2 between-subjects frame for both domains.

Next, the two-question clt has a measure for working memory (Digit span test) and inhibitory control (Stroop task). Finally, to trace the probability weighting function (pwf), we use a variation of the common consequence ladder (ccl) game. A typical ccl design requires participants to choose between successive risky (R) or safe (S) pairs related to each other through equivalent probability manipulation. The two prospect lists differ only in one aspect – decision attribute. The subjects in each state face identical five lottery pairs that differ in payoff-use case – for the same period (Decision Card Y) and six months hence (Decision card Z). This second intervention gives the within-subject measure of tunnelling. The comparisons indicate the degree of realignment, thus testing the primary hypotheses for the two domains. Thus, each choice is a function of the expected, unexpected scarcity and relevance/irrelevance status.

Overall, the design has two cross-cutting treatment arms - Scarcity and Decision Attribute. The first varies the nature of scarcity for treatment and control. Once the 2*2 (Natural*Scarcity treatment) states are fixed for all participants, the treatment of decision relevance is introduced, making it a 2*2*2 design. The experiment is a factorial combination of between and within-subject design and measures the mechanism behind the causality – the clt.
Randomization Method
Coupon system
Randomization Unit
Individual level randomisation for domain and scarcity treatment.
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
Individual-level clustering at experiment nodes. Total participants across the two periods = 400+400.
Sample size: planned number of observations
Same as cluster
Sample size (or number of clusters) by treatment arms
In all, we have the following treatments - domain (gains, losses), scarcity (natural, primed) and decision attribute (relevant, irrelevant).
The first two are assigned between subjects, while the latter is within. Therefore, once the domain and scarcity treatment level is fixed, all participants face the second treatment.
Sample size by treatment arms -
Period 1 (Natural scarcity) = 400
Gains + Losses = 200 +200
Gains in period 1 = 200
Break up of 200 participant sample in the gains domain in period 1 -
full treatment = (natural x primed x decision attribute) = 100
natural x non-primed x decision attribute treatment = 100
Losses in period 1 = 200
Break up of 200 participant sample in the losses domain in period 1 -
full treatment = (natural x primed x decision attribute) = 100
natural x non-primed x decision attribute treatment = 100

Period 2 (No natural scarcity) = 400
Gains + Losses = 200 +200
Gains in period 2 = 200
Break up of 200 participant sample in the gains domain in period 2 -
no natural scarcity x primed x decision attribute = 100
no natural scarcity x non-primed x decision attribute treatment = 100
Losses in period 2 = 200
Break up of 200 participant sample in the losses domain in period 2 -
no natural scarcity x primed x decision attribute = 100
no natural scarcity x non-primed x decision attribute treatment = 100
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Running 1000 simulations with simes multiple hypotheses correction, we find a 79.7% power for our primary hypothesis. That is, given a sample size of 400, of the thousand samples of such kind, we can reject our null of no tunnelling 79.7% of the time. The standard deviation is 0.097
Supporting Documents and Materials

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Institutional Review Boards (IRBs)

IRB Name
University of East Anglia
IRB Approval Date
IRB Approval Number


Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information


Is the intervention completed?
Data Collection Complete
Data Publication

Data Publication

Is public data available?

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials