“How Saudi Customers Actually Choose” — study protocol

• الحالة / Status: مخطط لها · planned. لا توجد بيانات ولا نتائج. / No data, no results.
• الإصدار / Version: الإصدار 1.0 · أكتوبر 2026 · الحالة: مخطط (المنهجية أولًا) / Version 1.0 · October 2026 · Status: planned (method first)
• القاعدة / Rule: المنهج وحجم العيّنة يُنشران أولًا، ثم النتائج. / Method and n are published first, then results.
• المسؤول / Owner: سيف الدين الحاج · Seifelden Alhajj. وجهة التنفيذ والمراجعة الأخلاقية لم تُحدّد بعد (انظر open_questions.md). / The executing entity and ethics review are not yet decided (see open_questions.md).

1. Question

What moves consumer choice in the Kingdom between realistic offers, in three everyday commercial decisions, and what do people give up for what?

The study measures trade-offs between attributes, not "what they'd pay". Riyal translations come with their limits (see 9).

2. Three independent studies

Levels are built from prices and times that actually exist in the market, collected and documented before design. No level a consumer has never seen is used.

3. Design

• At most six attributes per study, and at most five to seven levels for price (Sawtooth guidance).
• 10 to 12 tasks per respondent, three alternatives each, and a dual-response "none" (Brazell et al. 2006).
• An efficient, level-balanced design with no dominated option and minimal overlap (Huber & Zwerina 1996). Several questionnaire versions are assigned at random.
• Two fixed holdout tasks, excluded from estimation, measure predictive validity.
• A cheap-talk warning before the start (Cummings & Taylor 1999).
• Incentive alignment in S3 where legally and practically feasible: a draw in which the winner receives the pack the model picks for them (Ding et al. 2005). Competition rules are reviewed first (open question).

4. Sample

• Minimum: Johnson's rule as reported by Orme: n·t·a/c ≥ 500, ideally ≥ 1,000. For S1, with t = 12, a = 3 and c = 4, the theoretical minimum is 56 and the preferred figure 112.
• Adopted size: at least 300 respondents per study, and at least 200 per segment compared (e.g. Riyadh vs Jeddah, with vs without children). Expect 600–800 per study if two or three segments are compared.
• Eligibility: resident in the Kingdom, 18 or older (see 6), and the decision-maker or co-decider (e.g. ordered delivery in the last 30 days, for S1).
• Quotas: city, gender, age and nationality (Saudi or resident), under a frame fixed at pre-registration. National representativeness is not claimed unless the sampling frame supports it.

5. Recruitment and instrument

• Channel: a licensed online panel or research partner. The provider, recruitment method and incentives are recorded in the pre-registration.
• Language: Arabic first, English optional. Written in plain spoken Arabic, then language-reviewed.
• Comprehension testing: 20 think-aloud interviews before launch, then a soft launch of about 50 respondents to check duration and straight-lining.
• Data quality:
• A respondent is excluded for falling under a pre-set minimum duration, answering in a fixed pattern, or failing one pre-specified attention check.
• The exclusion rate is published.

6. Privacy and the PDPL

The reference is the Personal Data Protection Law (Royal Decree M/19 of 9/2/1443 AH, amended by M/148) and its Implementing Regulation. The Arabic text prevails, and legal counsel reviews the study before launch.

• Consent: explicit consent before any question, withdrawable at any time (Art. 5).
• Notice at collection: covers the controller, purpose, legal basis, mandatory and optional fields, retention period, how to withdraw consent, and rights (Arts. 12–13; Regulation Art. 4).
• Data minimisation: Art. 11(3).
• No name, mobile number or email in the analysis file.
• If recruitment or incentives need contact details, the recruiter holds them in a separate file the analyst never receives.
• No sensitive data (health, belief, origin, etc.).
• Research purposes: if the Art. 27 route is used, the duties in Regulation Art. 30 apply:
• record the purpose;
• collect the minimum;
• pseudonymise where possible;
• avoid negative impact on the data subject.
• Anonymisation:
• Published aggregated data is irreversibly anonymised. Anonymised data is no longer personal data (Regulation Art. 9).
• Pseudonymised data remains personal data.
• Retention and destruction: individual data is destroyed once the purpose ends (Art. 18), with the period stated in the notice.
• Transfer outside the Kingdom: in-Kingdom hosting is preferred. Any foreign platform is reviewed against Art. 29 and the amended 2024 Transfer Regulation.
• Breach: notify the competent authority within 72 hours (Regulation Art. 24).
• Minors: no participants under 18. The law sets no age itself, so this is a precautionary choice. The ICC/ESOMAR Code defines a child as 12 and under.
• Register and DPO:
• Registration on the national controllers register is not automatically required for a study of this kind with no sensitive data.
• A DPO is not automatically required either.
• Both are to be confirmed with counsel.
• Ethics:
• The 2010 Law of Ethics of Research on Living Creatures is written mainly for biomedical research; its application to commercial surveys is not explicit.
• If a university or research body takes part, the study goes to a local committee registered with the NCBE.
• In every case the ICC/ESOMAR Code (2025 revision) is followed.

7. Pre-registration

Before any data is collected, the following are registered on OSF Registries (embargo up to four years) or AsPredicted:

• questions and hypotheses;
• attributes, levels and their sources;
• design and number of versions;
• sample size and exclusion rules;
• analysis model;
• validity measures.

Any later deviation is declared and justified.

8. Analysis

• Models: aggregate conditional logit, then hierarchical Bayes for individual weights, and latent class for segments, with the number of classes chosen by a pre-specified criterion.
• Validity:
• Holdout hit rate against chance (25% with four options).
• Simulator shares compared with real sales where a partner allows, with the calibration documented (Orme & Johnson 2006).
• Willingness to pay: computed inside a market simulator against defined rivals, with uncertainty intervals. No standalone WTP figure is published without that context (Orme 2001).
• Tools: open source where possible, with code published:
• R: mlogit, logitr, cbcTools, idefix, bayesm, apollo;
• Python: biogeme, xlogit, choice-learn.
• A commercial tool such as Sawtooth or Conjointly may be used; naming one is not an endorsement.

9. Publication policy

1. First, the method: the registered protocol, the sample size, the full Arabic instrument, and response and exclusion rates.
2. Second, results with their limits: - weights and importances within the tested ranges only; - simulations against defined scenarios; - uncertainty intervals; - an explicit warning that share of preference is not market share and that riyal figures tend to overstate.
3. Third, the data: aggregated, anonymised data and the code.

No result is used to market services before the method is published. Nothing is attributed to "the Saudi customer" beyond what the sampling frame supports.

10. What the study does not promise

No market "average", no market price elasticity, and no generalisation beyond the tested categories. Cost and schedule are not published here until the executing entity and actual quotes are known.

Back to the paper