# Data dictionary: South Australian poll, August 2026

This dictionary describes the two files published with *Back to the line* at acestrategies.au/intelligence/back-to-the-line/. The CSV `sa-poll-series.csv` holds one row per reading. The JSON `sa-poll-august-2026.json` holds every figure the page draws. Every figure in both is DemosAU's as published in its South Australian reports for June 2026 (fieldwork 29 May – 15 June 2026) and August 2026 (fieldwork 5–26 August 2026), with the 21 March 2026 election result as DemosAU reports it. Page numbers refer to the August 2026 report unless a field says otherwise.

## How an absent figure is written

| Where | An absent figure is | Meaning |
|---|---|---|
| CSV | an empty cell | The reading has no such figure. On the election row, the sample, the effective sample and the margin of error are empty because the election is a count rather than a survey. An empty cell never means zero. |
| JSON | a missing key | The report does not publish that figure for that reading. A key is never set to null or to zero to stand for a missing figure. |
| JSON, vote flows | a row that sums to less than 100 | The remainder is movement the report does not itemise. It is not a party, and the page draws it in grey as unlabelled. |

## The CSV: one row per reading

The CSV has one row for the election and one for each poll, in date order, with a header row.

### `wave`

- **Type:** string
- **Unit:** none
- **Source:** the report's own label for the reading
- **Note:** The label is "Election 21 Mar 2026", "June 2026" or "August 2026". It is never empty.

### `kind`

- **Type:** string
- **Unit:** none
- **Source:** Ace Strategies
- **Note:** The value is `election` for the count and `poll` for a DemosAU wave. It is never empty.

### `fieldwork_start`

- **Type:** date
- **Unit:** ISO 8601, YYYY-MM-DD
- **Source:** the report's methodology statement (p. 2)
- **Note:** This is the first day of fieldwork. On the election row it is polling day, 2026-03-21.

### `fieldwork_end`

- **Type:** date
- **Unit:** ISO 8601, YYYY-MM-DD
- **Source:** the report's methodology statement (p. 2)
- **Note:** This is the last day of fieldwork, and it is never earlier than `fieldwork_start`. On the election row it is polling day.

### `n`

- **Type:** integer
- **Unit:** respondents
- **Source:** the report's methodology statement (p. 2)
- **Note:** This is the sample DemosAU published. It is empty on the election row.

### `effective_n`

- **Type:** integer
- **Unit:** respondents
- **Source:** the report's methodology statement (p. 2)
- **Note:** This is the effective sample after rim weighting, on which the margin of error is computed. It is empty on the election row.

### `moe`

- **Type:** number
- **Unit:** percentage points, plus or minus
- **Source:** the report's methodology statement (p. 2)
- **Note:** This is the margin of error on the effective sample at the 95 per cent level. It applies to the state-wide figures only, and every subgroup figure carries a wider margin. It is empty on the election row.

### `ALP`, `ONP`, `LIB`, `GRN`, `OTH`

- **Type:** number
- **Unit:** per cent of the primary vote, undecided excluded
- **Source:** the report's voting-intention tables (pp. 3–4)
- **Note:** These are the primary votes for Labor, One Nation, the Liberals, the Greens and any other candidate. The poll rows are whole numbers as published, and the election row carries one decimal place. Each row sums to 100.

### `tpp_ALP`, `tpp_ONP`

- **Type:** number
- **Unit:** per cent of the two-party-preferred vote, Labor against One Nation
- **Source:** the report's two-party table (pp. 3–4)
- **Note:** These are DemosAU's two-party estimates. The two figures in a row sum to 100.

### `tpp_basis`

- **Type:** string
- **Unit:** none
- **Source:** DemosAU
- **Note:** This names how the two-party figure was built. "DemosAU estimate" marks the election baseline, which applies DemosAU's March 2026 flow estimates to the state-wide primary vote. "DemosAU 2026 flow estimates" marks a poll figure built from the same flows.

## The JSON: every figure the page draws

| Key | Type | Unit | Source | What it holds |
|---|---|---|---|---|
| `series` | string | none | Ace Strategies | This is the name of the series. |
| `jurisdiction` | string | none | Ace Strategies | This is the state code, `SA`. |
| `units` | string | none | Ace Strategies | This states the unit every voting-intention figure is in. |
| `parties` | object | none | DemosAU | This maps each party key (`ALP`, `ONP`, `LIB`, `GRN`, `OTH`) to its name in the question. |
| `leaders` | object | none | DemosAU | This maps each party key to its leader and the leader's office. |
| `preference_flows_2026_estimate` | object | per cent | DemosAU (method note) | This holds, for Liberal, Greens and other voters, the share of preferences flowing to Labor (`ALP`) and to One Nation (`ONP`), with a `source` string. |
| `waves` | array | none | see below | This holds one object per reading, in date order: the election, June 2026 and August 2026. |

### `waves[]`

| Field | Type | Unit | Source | What it holds |
|---|---|---|---|---|
| `id` | string | none | Ace Strategies | This is a slug unique in the file, such as `demosau-2026-08`. |
| `kind` | string | none | Ace Strategies | This is `election` or `poll`. |
| `label` | string | none | Ace Strategies | This is the reading's display label. |
| `date` | date | ISO 8601 | ECSA | This is polling day, and only the election carries it. |
| `source` | string | none | DemosAU | This names where the election figures come from, and only the election carries it. |
| `fieldwork.start`, `fieldwork.end` | date | ISO 8601 | p. 2 | These are the first and last days of fieldwork, and only a poll carries them. |
| `n` | integer | respondents | p. 2 | This is the published sample. |
| `effective_n` | integer | respondents | p. 2 | This is the effective sample after weighting. |
| `moe` | number | points | p. 2 | This is the margin of error on the effective sample. |
| `undecided_excluded` | number | per cent | pp. 3–4 | This is the share still undecided after the leaner question, excluded from the voting-intention totals. |
| `primary` | object | per cent | pp. 3–4 | This holds the primary vote by party key. |
| `tpp` | object | per cent | pp. 3–4 | This holds the two-party estimate, `ALP` and `ONP`, with a `note` on the election. |
| `preferred_premier` | object | per cent | DemosAU report | This holds each option's share, the `options` offered in that wave, and a `note` where the options changed. The June and August questions offered different options, so the two waves are not a series for this measure. |
| `leader_ratings` | object | per cent and points | DemosAU report | This holds, for each leader, the positive (`pos`), neutral (`neu`) and negative (`neg`) shares and the published net rating (`net`), which is positive less negative. |
| `direction` | object | per cent | DemosAU report | This holds the right-direction (`right`), wrong-direction (`wrong`) and don't-know (`dk`) shares. |
| `biggest_issue` | object | per cent | DemosAU report | This holds each open-ended category's share. The category sets differ between waves, and only cost of living and housing are compared across them. |
| `cohorts` | object | per cent | DemosAU report | This holds the primary vote by party key for each published cohort: gender, age, education, housing tenure and region. Every cohort figure carries a margin wider than `moe`. |
| `march_vote_flows` | object | per cent of the row | p. 10 (June report p. 10 for June) | This holds, for each March 2026 vote, the share now intending each party. A row that sums to less than 100 carries movement the report does not itemise, and a `note` records where the report prints a segment without a party label. |

## Derived figures

| Derived figure | How it is built |
|---|---|
| Change since the election and since June | August primary vote less the election result, and less June, in percentage points, by party. |
| Retention | The share of each party's March voters still intending that party: the party's own cell in `march_vote_flows`. |
| Scenario two-party (projection) | Labor primary plus the Liberal, Greens and others primaries each multiplied by its share flowing to Labor. At the default flows on the August primaries it gives 57.8, and the report's 58 is computed on unrounded shares. |
| Days since the election | The days from polling day to the close of fieldwork. |

## Citing a figure

A figure from these files is cited as DemosAU, South Australian poll, with its fieldwork dates, published by Ace Strategies. A derived figure is cited as Ace Strategies analysis of DemosAU figures, and a scenario figure is cited as a projection.