Brand tracker study planner
How often should you measure your brand?
Compare tracking schedules against the change you need to detect, the audiences you need to hear from, and the budget you can commit.
Start with the decision, then test the schedule.
Plan independent, repeated cross-sectional surveys with percentage-based measures. Compare seven collection and reporting options using an annual interview limit, your own cost estimates, or a required change-detection target.
Editable example: 2,400 usable interviews a year, a quarterly decision cycle, and a 10-point change.
Model-based study planning
Candidate: Quarterly waves
This option meets the entered change target for both metrics within your decision cycle. The rule favors the longest suitable period, then lower entered cost, then fewer collection periods.
This is a planning candidate. Timing, feasibility, and continuity still need review. The selection below is the option you are inspecting; it can differ from the candidate.
Unallocated interview capacity: 0. Collection periods use equal whole-person targets.
The base for your priority measure
Expected natural bases; group incidence can vary. These are not quota guarantees.
Sample source is opt-in or unconfirmed. These are model-based precision references, not a population sampling margin of error.
Overall measure Baseline 50%
- Planning effective base
- 600
- 95% single-period precision
- ±4 points
- Detectable increase at 80% power
- 8.05 points
- Power for your 10-point increase
- 93.7%
Priority measure Baseline 50%
- Planning effective base
- 600
- 95% single-period precision
- ±4 points
- Detectable increase at 80% power
- 8.05 points
- Power for your 10-point increase
- 93.7%
Change tests compare equal independent periods against zero, using a two-sided 5% significance threshold per comparison. Single-period precision does not measure change sensitivity. An unavailable detectable change is not zero.
A schematic first year
One window each quarter. Reports: separate quarterly waves.
F = collection; R = report; E = marked event; M = module. Wave months are illustrative. Report markers omit processing time; agree actual fieldwork dates and delivery lag.
- Jan– –
- Feb– –
- MarF R
- Apr– –
- May– –
- JunF R
- Jul– –
- Aug– –
- SepF R
- Oct– –
- Nov– –
- DecF R
400 usable-participant hours annually at 10 core minutes and 0 extra minutes per module interview. Rejected interviews and screening time are excluded.
Annual cost from your assumptions
Add your cost estimates to compare setup, fieldwork, gross completes, reporting, and modules. No supplier rate is assumed.
Before comparing the trend
Trend continuity is unconfirmed. Check population, sample source, wording, mode, weighting, timing, exclusions, and respondent reuse before interpreting movement.
Your entries are calculated in this browser. Copy and download include the selected option, alternatives, sensitivity, assumptions, and sources.
Compare the tracking options
Same inputs, different schedules. Read power beside your 10-point meaningful change. Each option uses its own annual overhead in budget mode.
| Option / reporting | Per report | Priority base | Power: overall / priority | Annual usable | Annual cost | Cadence and power |
|---|---|---|---|---|---|---|
| Separate months | 200 | 200 | 52% / 52% | 2,400 | Not entered | Review sample / method |
| Calendar quarters | 600 | 600 | 93.7% / 93.7% | 2,400 | Not entered | Meets entered criteria |
| Calendar quarters | 600 | 600 | 93.7% / 93.7% | 2,400 | Not entered | Meets entered criteria |
| Trailing three months | 600 | 600 | Withheld / Withheld | 2,400 | Not entered | Overlap method needed |
| Separate quarterly waves | 600 | 600 | 93.7% / 93.7% | 2,400 | Not entered | Meets entered criteria |
| Separate half-year waves | 1,200 | 1,200 | 99.9% / 99.9% | 2,400 | Not entered | Slower than decision cycle |
| Yearly | 2,400 | 2,400 | >99.9% / >99.9% | 2,400 | Not entered | Slower than decision cycle |
Annual change needs a second independently sampled year. Rolling reports share respondents across adjacent windows. Calendar-quarter aggregates do not overlap if people appear once. Meeting these numeric criteria does not validate the sample source or trend continuity.
What changes the selected option’s feasibility?
Specified sensitivity scenarios, not confidence bounds. Required-sample mode recalculates sample targets; constrained modes keep the constraint unless the row changes it.
| Assumption | Annual usable | Priority effective base | Priority power | Annual cost |
|---|---|---|---|---|
| Current assumptions | 2,400 | 600 | 93.7% | Not calculated |
| Design effect 1 | 2,400 | 600 | 93.7% | Not calculated |
| Design effect 1.5 | 2,400 | 400 | 81.3% | Not calculated |
| Design effect 2 | 2,400 | 300 | 69.3% | Not calculated |
| Half the meaningful change | 2,400 | 600 | 41.1% | Not calculated |
| No audience / metric filtering | 2,400 | 600 | 93.7% | Not calculated |
| 25% more annual interviews | 3,000 | 750 | 97.4% | Not calculated |
Calculation conventions and your study specification
Equal collection allocations; conditional shares multiplied in sequence; effective base = usable base / design effect. Gross completes = ceiling(usable / (1 − rejection)) per collection period. Precision = normal critical value × √(p(1 − p) / effective base). Independent percentage power uses pooled variance under the null and separate variance under the alternative.
Expected yes/no counts below 10 withhold affected estimates. Numerical sample searches stop at one million usable people per collection period; a failed search is reported. Rolling windows, paired people, and unsupported measures receive no substitute independent-wave power. Full policy and sources are below.
Fixed planning example; interactive controls require JavaScript.
Collection, reporting, and decisions have different schedules.
When people answer
One monthly fieldwork window and interviews spread throughout a month can produce the same count. They capture different timing. The continuous option uses monthly operating periods for budgeting; it does not promise precise daily results.
Which answers make a report
Separate monthly reads, calendar-quarter aggregates, and rolling quarters use different windows. Adjacent rolling quarters share two months of answers. Their change cannot use the independent-wave formula.
When the business can respond
A monthly report is useful only if it supports a decision. The planner checks the period length against your action cycle and the sample against your change target. Category dynamics and delivery time still need judgment.
Keep the denominator visible.
A survey of 600 people can leave only 84 able to answer a question after filtering to 70% category buyers, 40% within an age group, and 50% who know the brand. Each rate is conditional on the preceding group. Weighting can reduce the planning effective base further.
Natural group shares are uncertain. If a particular group must support a decision, discuss a quota or oversample and its analysis implications. This release evaluates one priority group; it does not optimize several overlapping quotas.
Explore brand tracking researchDoes a five-point target mean five points will be significant?
No. Power is the probability of detecting an assumed true change under repeated studies. A five-point target is not a significance cutoff, and power to detect it does not prove the change exceeds five points. This planner tests against zero and does not analyze observed results.
What should we ask a sample supplier?
Ask how people are recruited, whether sources are blended, how that mix is controlled between waves, how duplicate or fraudulent responses are removed, and whether quotas or weighting targets can change.
Confirm the recontact policy and whether anyone can appear in both periods being compared. A panel can supply fresh people for each comparison, but panel membership alone does not establish independence.
Can we add a campaign module?
Mark the relevant months and enter extra interview time and a cost per module window. The schedule flags event months without collection. Modules do not add respondents or improve the core metric’s sample. Temporary sample boosts need separate allocation and costing; movement around a campaign does not establish campaign causation.
How do we preserve the existing trend?
Document population definitions, source mix, eligibility, quotas, weighting, wording, order, mode, timing, and exclusions. If a material method changes, assess parallel old/new measurement or a documented break. A larger sample cannot correct a measurement discontinuity.
Methods and research
A disclosed planning model, with explicit limits.
Percentage precision: a normal critical value multiplied by √(p(1 − p) / effective base). It is a single-period reference under the entered assumptions, not a measure of total survey accuracy. A design effect divides the usable base for sensitivity; it does not reconstruct the actual sample design.
Independent-wave change: the existing percentage-power engine uses pooled variance under the null, separate variance under the alternative, and both rejection tails. There is no continuity correction. Numerical searches find sample or detectable change. The optional Bonferroni count divides the family significance threshold; power is per comparison, not joint power for the whole tracker.
Bounds and sparse data: the selected baseline and change must leave each metric strictly between 0% and 100%. Expected yes/no counts below 10 withhold affected estimates. Searches stop at 1,000,000 usable interviews per collection period. Sparse outcomes, repeated people, NPS, and overlapping-window change require another method. Even adequate counts leave approximation error.
Operations: equal whole-person collection targets; expected conditional bases retain fractions; gross completed interviews round up separately per period after the entered quality loss. Annual costs sum entered setup, collection, gross interviewing, reporting, modules, and other costs. No screening-yield model, supplier rate, recruitment guarantee, or campaign lift is inferred.
The candidate rule is a disclosed planning convention, not a validated predictor of the best tracking design. This release does not model historical trends, paired respondents, seasonal effects, several markets, optimized subgroup boosts, or causal effects. Its download preserves the assumptions and exclusions. Sources reviewed September 29, 2026:
- Statsmodels: power for two independent proportions
The pooled-null, unpooled-alternative normal power calculation used here.
- NIST: difference of proportions
Independent samples are required for the ordinary two-proportion comparison.
- AAPOR: online sample quality report
Pages 38–40 discuss uncertainty, unequal weighting, and the limits of simple-random-sample formulas.
- AAPOR: best practices for survey research
Supports consistent measurement across time and controlled assessment of methodology changes.
- AAPOR: 2026 survey costs publication
Costs cover more than interviewing; disclose included and excluded components.
- Ipsos: get the tracking frequency right
Practitioner guidance connects cadence to category activity and the ability to act, rather than a universal frequency.