How the Wave 1 survey was conducted

1. Introduction and research objectives

The CDD National Opinion Tracker is an independent, non-partisan public opinion research initiative undertaken by the Centre for Democracy & Dialogue (CDD). It is designed to provide systematic, evidence-based information on voting intentions, political leadership preferences and the issues influencing electoral choices in Fiji.

The Tracker aims to contribute to informed public debate by providing regular measurements of public opinion using transparent research methods. It is intended to operate through successive survey waves, allowing changes in public attitudes and electoral preferences to be monitored over time.

The first wave was conducted during September and October 2026 and comprises 1,673 completed interviews with eligible voters across Fiji's four administrative divisions.

The survey employed a quantitative, cross-sectional research design using stratification by geographic division and ethnicity, software-assisted random respondent selection, standardised interviewing procedures and post-survey statistical weighting.

The findings provide a snapshot of respondents' expressed preferences during the survey period. They should not be interpreted as predictions of the next General Election's outcome.

2. Target population and geographic coverage

The target population comprised individuals who:

• Were citizens of Fiji;
• Were aged 18 years or older;
• Were eligible to vote in the next General Election;
• Ordinarily resided in Fiji; and
• Voluntarily consented to participate.

The survey covered all four administrative divisions: Central, Eastern, Northern and Western.

The final analytical sample consisted of 1,673 completed interviews. Of these, 1,360 respondents identified a political party or independent candidate they would vote for, while the remaining 313 were undecided, would not vote, or declined to identify an electoral preference.

All completed interviews in the final dataset are recorded as telephone interviews.

3. Sampling design and rationale for stratification

A central consideration in the design of the CDD National Opinion Tracker was the need to account for the continued significance of ethnicity in Fiji's electoral politics.

Fiji's voting patterns have historically been strongly associated with ethnic identity. While political preferences are also shaped by leadership, economic circumstances, public policy, geography and other factors, ethnicity remains an important variable associated with voting intention.

For this reason, CDD considered it methodologically important that the sample adequately represent Fiji's major ethnic communities, both nationally and within the country's different geographic divisions.

The ethnic composition of the population varies considerably between divisions. A sample reflecting Fiji's overall national ethnic composition could therefore still be geographically unbalanced. For example, an appropriate national proportion of iTaukei respondents could be achieved while drawing too many from Central Division and too few from Western or Northern Division.

To address this, CDD adopted a stratified sampling structure based on the intersection of ethnicity and administrative division. Four divisions were combined with three broad ethnic categories:

• iTaukei;
• Indo-Fijian; and
• Other ethnicities.

This produced 12 sampling strata.

The use of these strata was intended to improve demographic and geographic coverage and ensure that no ethnic community or geographic area exerted disproportionate influence over the published national results because of its representation in the achieved sample.

The adoption of ethnic stratification should not be interpreted as assuming that members of the same ethnic community hold identical political views. Rather, it acknowledges the importance of ethnicity as a variable in Fiji's political landscape and the need to account for its geographic distribution.

4. Population data and determination of sampling targets

CDD obtained demographic information from the Fiji Bureau of Statistics (FBOS) to establish the relative population sizes of the 12 strata. The First Wave Sampling and Fieldwork Guide also used registered-voter data to establish broad divisional recruitment targets. Population benchmarks were used to guide the allocation of respondents among divisions and ethnic categories.

The original fieldwork plan provided for 800 completed interviews. It deliberately allocated a larger proportion of interviews to Eastern Division than that division's population share, to improve the scope for examining results from that relatively small population. The final achieved survey sample comprised 1,673 respondents.

For statistical adjustment of the achieved sample, the Tracker used division-by-ethnicity population benchmarks derived from the 2017 FBOS census data, adjusted to approximate the population aged 18 and above.

Because the census age tables contain a 15–19-year age category, the estimated population aged 18–19 was calculated as two-fifths of that five-year category. This approach assumes an approximately even distribution of ages within the bracket and is therefore an approximation.

The demographic benchmarks represent estimates of the adult resident population rather than a direct count of registered or eligible voters. This distinction is recognised as a limitation.

5. Construction of respondent-selection pools

CDD placed particular emphasis on minimising selection bias during recruitment. Rather than allowing enumerators to select individuals based on convenience, availability, personal acquaintance or perceived political preferences, the research team constructed a pool of potential respondents for each of the 12 strata.

Each pool was designed to contain at least twice the number of eligible individuals required to meet the target sample size for that stratum. This approach ensured that the research team had access to a larger pool of potential participants than the number required for completed interviews. It also provided scope for managing refusals, unsuccessful contacts and other recruitment difficulties without automatically relying on interviewer-selected substitutes.

A software application was used to randomly select potential respondents from each stratum-specific pool. The selection process was designed to minimise interviewer discretion and prevent subjective decisions about which individuals should participate.

Importantly, the random-selection process was undertaken independently of respondents' political preferences. Individuals were not selected, excluded or replaced because of support for a particular political party, leader or political position.

Random selection within a constructed pool provides an important safeguard against interviewer selection bias. However, the overall representativeness of the sample also depends on how those pools were assembled and whether relevant sections of the electorate were adequately covered.

6. Non-response, refusals and replenishment of selection pools

One of the most significant challenges encountered during fieldwork was respondents' reluctance to discuss their political preferences. A substantial number of individuals selected through the random-selection software declined to participate after learning that the survey concerned their voting intentions and political choices. Interviewers encountered hesitation, apprehension and distrust among some potential respondents.

These experiences suggested that discussing political preferences remains a sensitive matter for parts of Fiji's population. Participation was entirely voluntary, and individuals who declined were not pressured to participate.

These refusals created an operational challenge. As randomly selected individuals declined, the available pool of potential respondents within some strata diminished. CDD therefore replenished the relevant selection pools to maintain a sufficiently large group of potential respondents from which the software could continue selecting randomly.

The objective was to avoid a situation in which the research team exhausted the original selection pool and subsequently relied on whichever individuals happened to be available or willing to participate. Pool replenishment was consequently an important part of the process for preserving software-assisted random selection during fieldwork. However, identifying additional eligible respondents, rebuilding the selection pools and undertaking further selection rounds increased the time required to complete the survey.

Although this approach reduced reliance on convenience recruitment, it could not eliminate non-response bias. Individuals who willingly disclose political preferences may differ in important ways from those who decline to participate. This possibility must be considered when interpreting the findings.

7. Interviewer training and questionnaire administration

Enumerators were trained in the principles of impartial research, respondent confidentiality, neutral questioning and accurate recording of responses. The questionnaire was designed to obtain responses without suggesting particular political parties, leaders or issues. Respondents were informed about the identity of CDD, the purpose of the research, the voluntary nature of participation and the confidentiality of their answers.

The core survey covered:

• Voting intention;
• Certainty of electoral choice;
• Preferred Prime Minister;
• The most important issue influencing voting decisions; and
• Demographic characteristics.

The voting-intention question asked:

"If the General Election were held tomorrow, which political party would you vote for?"

Enumerators were specifically instructed not to read the list of political parties unless respondents requested the available options. Responses were to be recorded as provided. Only respondents who identified a political party or independent candidate were asked whether their electoral preference was final or whether they could still change their minds.

The preferred-Prime-Minister question was open-ended. Enumerators were not permitted to suggest names. Similarly, the question concerning the most important issue influencing voting decisions was open-ended, with no predetermined list of issues read to respondents. This was intended to allow respondents to identify their own priorities without being influenced by suggested categories.

8. Observations regarding political party recognition

An important observation made during fieldwork was that some respondents appeared unfamiliar with the names of political parties. In certain cases, respondents were more familiar with individual political personalities than the parties associated with them. This presented a potential risk of interviewer influence, particularly where respondents sought clarification or referred to politicians rather than parties.

Enumerators were therefore trained to avoid interpreting an expressed preference for a political leader as support for a particular party, or otherwise directing respondents towards a specific electoral choice. The distinction between preference for a political party and preference for an individual leader is particularly important in the interpretation of the Tracker's results.

It must nevertheless be acknowledged that the survey did not include a separate political knowledge test. Consequently, these interviewer observations cannot be used to establish the proportion of voters who were unaware of party names, misunderstood political affiliations or confused one party with another. Such questions could be examined in future survey waves.

9. Data processing and statistical weighting

Following data collection, the achieved sample was assessed against the demographic benchmarks established for the 12 division-by-ethnicity strata. Statistical weighting was applied to adjust for differences between the achieved sample and the estimated adult-population distribution.

The weight assigned to respondents within each stratum was calculated as:

For example, the Central iTaukei stratum accounted for approximately 28.48% of the estimated adult population but 26.36% of the achieved sample. The adjustment factor for this stratum was therefore approximately 1.0803. Conversely, Eastern iTaukei respondents represented a larger proportion of the achieved sample than their population benchmark and received an adjustment factor of approximately 0.6381. The resulting weights were applied to produce population-adjusted percentages.

Wave 1 weighting factors

The weighting model adjusted for ethnicity and division jointly. Gender was monitored during fieldwork but was not independently calibrated in the final weighting model. Age and urban–rural residence were likewise not included as additional calibration variables.

It is important to recognise that weighting cannot eliminate every form of sampling or non-sampling error. In particular, it cannot fully correct non-response bias, inadequate coverage of potential respondents or inaccuracies in information provided during interviews.

10. Data analysis and presentation

Survey responses were analysed using weighted descriptive statistics. Voting-intention results are presented using two distinct respondent bases.

The first comprises respondents who identified a political party or independent candidate. This is referred to as the decided-voter base, comprising 1,360 respondents before weighting. The second comprises all 1,673 respondents, including undecided voters, those who would not vote and those who declined to identify a political preference. This distinction is important because percentages calculated among decided voters will differ from percentages calculated among all respondents.

The survey also examines differences in responses by ethnicity, division, gender, age and urban–rural residence. Subgroup analysis is treated with caution, especially where the underlying number of respondents is small.

Responses concerning preferred Prime Minister were standardised to account for spelling variations and different forms of personal names. Open-ended responses concerning important electoral issues were classified into analytical themes. Some answers mentioned more than one issue, despite respondents being asked to identify a single most important issue. For that reason, issue-mention percentages may overlap and should not automatically be expected to total 100%.

Results derived from open-ended responses require consistent coding rules and appropriate review of ambiguous answers.

11. Methodological limitations

11.1 Non-response bias and political sensitivity

A significant number of potential respondents declined participation after learning that the survey concerned political preferences. Their reluctance may have reflected privacy concerns, distrust, political apprehension or other considerations. If those who refused participation differ systematically in their political preferences from those who agreed, the final sample may contain non-response bias. Random selection and demographic weighting cannot fully eliminate this risk.

11.2 Coverage of respondent-selection pools

Although respondents were randomly selected by software, the approach depended upon the composition of the pools from which selection occurred. Voters who were absent from those pools may have had little or no opportunity to participate. Consequently, the overall sampling design should not be represented as giving every eligible voter an equal or known probability of selection unless the construction of the selection pools independently establishes that property.

11.3 Telephone coverage

All completed interviews in the final dataset are recorded as telephone interviews. This may limit representation of individuals without reliable telephone access, those living in areas with poor connectivity, people not included in available contact databases, and those unwilling to discuss politics over the telephone. The fact that CDD's general methodology permits face-to-face or other modes does not mean those methods were used in Wave 1.

11.4 Age distribution

The achieved sample contains a relatively high proportion of younger respondents. Respondents aged 18–34 account for 842 interviews, approximately 50.3% of the total sample. By contrast, only 49 respondents, approximately 2.9%, were aged 65 or older. Because age was not independently included in the weighting calibration, this imbalance could affect overall estimates if political preferences are associated with age. It also limits the reliability of findings specifically concerning older voters.

11.5 Urban and rural representation

Of the 1,673 respondents, 1,210 identified their usual residence as urban, representing approximately 72.3% of the achieved sample. Urban–rural residence was not independently calibrated through the final weighting model. This may introduce residual bias if the distribution of urban and rural respondents differs from the electorate and voting intentions vary by residence.

11.6 Small demographic strata

Some division-by-ethnicity strata contain very small numbers of respondents. For example, Eastern Division contains only two Indo-Fijian respondents and six respondents categorised as Other. While weighting adjusts their contribution to national estimates, it cannot make these small cells statistically robust. Results for such groups should not be published as reliable standalone subgroup estimates.

11.7 Age of demographic benchmarks

The population weights are based on demographic information derived from the 2017 census. Since then, migration, population ageing, changes in residence and other demographic developments may have altered the composition of Fiji's adult population. The weighting benchmarks also approximate the adult resident population rather than directly measuring the current registered electorate. These differences may affect the precision of the adjustment.

11.8 Political party recognition

Enumerators observed that some respondents did not appear to know the names of political parties. This could affect responses to the voting-intention question, especially where political preferences are associated primarily with particular personalities. Although enumerators were trained to minimise prompting and leading, the survey did not independently measure political knowledge. Therefore, no conclusions about the prevalence or causes of voter confusion can be established from this questionnaire alone.

11.9 Interviewer effects

Standardised questions and training were used to minimise interviewer influence. Nevertheless, tone, explanations, timing, language and interactions between interviewers and respondents may affect answers, particularly when discussing politically sensitive subjects.

11.10 Social desirability and reluctance to disclose preferences

Some respondents may have been unwilling to reveal their genuine political preferences, even after agreeing to be interviewed. Others may have provided answers they considered safer or more socially acceptable. Such effects are difficult to measure directly and may not be eliminated through interviewer training or statistical weighting.

11.11 Interpretation and coding of open-ended responses

The questionnaire collected open-ended answers concerning preferred Prime Minister and important voting issues. These responses required subsequent standardisation and classification. Coding involves methodological judgement and can introduce error, particularly where answers are unclear or mention several subjects. Quantitative analysis of these responses should therefore be accompanied by transparent coding rules and appropriate checks.

11.12 Sampling and statistical uncertainty

The relatively large overall sample does not mean that every estimate is equally precise. Uncertainty increases when examining small subgroups and narrow differences between parties. For reference, a simple random sample of 1,673 would have an approximate maximum theoretical sampling error of ±2.4 percentage points at the 95% confidence level, while a simple random sample of 1,360 would have a corresponding figure of approximately ±2.7 points. These are illustrative mathematical benchmarks only. They should not be described as verified margins of error for the Tracker because the actual sampling design, pool coverage and non-response introduce additional uncertainties.

11.13 Limitations of stated voting intentions

Survey respondents were asked about their intended choices at the time of the interview. Such preferences may change as a result of political events, leadership changes, policy announcements, economic developments or electoral campaigning. A respondent's stated intention does not guarantee that they will vote or make the same choice on polling day. The findings must therefore be treated as measurements of opinion during the fieldwork period rather than election forecasts.

11.14 Fieldwork duration and operational challenges

The reluctance of selected respondents to discuss political matters created additional recruitment demands. The replenishment of selection pools, further random selections and repeated recruitment efforts increased the time required to complete the survey. This experience demonstrates the practical difficulties associated with obtaining participation in politically sensitive research.

11.15 Data-record quality and reconciliation

Accurate fieldwork dates, respondent identifiers, geographic classifications and interview records are important to the integrity of the dataset. Where inconsistencies arise, they should be reconciled against source records before publication. Any material unresolved discrepancies should be disclosed, and the dataset used for weighting should be clearly identified and preserved.

12. Ethical standards, transparency and future improvements

CDD conducted the survey as an independent, non-partisan research exercise. Participation was voluntary, and respondents were informed of the confidentiality of their responses. The programme does not seek to promote or disadvantage any political party, candidate or political position.

The questionnaire has been made publicly accessible to allow interested parties to examine the wording and sequencing of questions. Future survey waves will seek to improve sampling coverage, reduce non-response, strengthen demographic balance, refine the coding of open-ended responses and maintain consistent statistical methods. CDD also intends to monitor differences between successive survey waves, while documenting any methodological changes that could affect comparability.

The limitations described above do not negate the value of the National Opinion Tracker. Rather, acknowledging them establishes the appropriate boundaries for interpreting its findings.

The objective of the CDD National Opinion Tracker is to provide an increasingly robust, transparent and independent body of evidence concerning public opinion in Fiji, while maintaining the scientific integrity and political neutrality necessary for credible electoral research.

See also: the Wave 1 public questionnaire.