In 2020, the Victorian Health Promotion Foundation (hereafter VicHealth) in consultation with policy, practitioner and academic stakeholders, commenced the development of a program to understand the impact of harmful industries on health and wellbeing outcomes in Australia. As a first step, VicHealth is piloting a model to systematically collate, analyse and document information on harmful industry activities in Australia, with a primary focus on the alcohol, gambling and ultra-processed food sectors, with some inclusion of tobacco. The first author (JLN) was employed to help scope, design and implement the pilot.

In consultation with an advisory group of public health researchers and NGOs, eight priority domains were selected for the pilot: lobbying, political contributions, revolving door, astroturf organisations, digital marketing, influencers, community sponsorship and corporate health promotion. Our methods for piloting a tool to monitor lobbying and political contributions are detailed here.

Our first step was to determine the availability of information available. For each political activity (lobbying and political donations), relevant datasets were identified and explored. The websites of the federal, state and territory governments were searched to identify records of ministerial diaries, lobbyist registers and political donation returns. While most jurisdictions provided lobbyist registers and records of political donations, only three states disclosed ministerial diaries. Table 1 documents the availability of datasets for each jurisdiction (updated 16/09/2021).

Two datasets were selected for the pilot: NSW ministerial diaries and the federal record of political donation returns.

Ministerial diaries from July 2014 – December 2020 (n = 598) were downloaded from the NSW government website (https://www.dpc.nsw.gov.au/publications/ministers-diary-disclosures/). Adobe Acrobat DC was used to convert PDF files to.xlsx (excel spreadsheet) files. Each file was manually cleaned and formatted in Excel (e.g., removed headers and footers, formatted columns, ensured dates were consistently formatted). The final table had 20,608 rows, each representing a unique meeting.

Political contribution data (including federal and state donations) was downloaded from the Australian Electoral Commission (AEC) Transparency Register (https://transparency.aec.gov.au/AnnualDonor) as a single.csv file and imported into Excel. For missing dates, original donor returns were reviewed. For returns that listed only the month (and not the specific date), then the first of the month was used for the date. For donations with no date listed (e.g., Star Entertainment $800 donation to Lib-NSW 2018–2019), the last day of the financial year was used, (e.g., 30/06/2019). In some cases, donations were presented as an annual aggregate of several donations (e.g., Motor Trades Electoral Action Committee 1998–1999 return). In this case, the date of the last donation was used. The original returns of eight donations listed as ‘zero’ dollars were all reviewed. The original donations were for amounts less than one dollar, and all made to the Citizens Electoral Council of Australia party. These were left as $0.00 to be consistent with AEC reporting.

Once we had collected all the data, our next step was to organise the data so that it could be analysed and visualised. To analyse the data, the lobbying and political contribution datasets were organised into fact tables (containing the data observations, e.g., a specific political donation) and dimension tables (which contain descriptive attributes about the variables in the fact table, e.g., the political party affiliation of the donation recipient). This approach follows Tidy Data principles, [45] which aims to make ‘messy’ datasets ‘easy to manipulate, model and visualise’ by applying a similar structure. This segmentation is important for data warehouse design, which allows you to build relationships between the fact and dimension tables, so that, for example, it is possible to filter political donations by the industry affiliation of the donor. While a data warehouse was not within the scope of the pilot, it is a logical next step, thus we applied data warehouse design principles to the organisation of our data [46].

For this project, two fact tables (one of ministerial meetings and a second of political donations) and four-dimension tables were created for coding purposes. The dimension tables corresponded to three variables in the fact tables: 1) harmful industry groups (alcohol, tobacco, gambling and ultra-processed foods) used to code the donors and meeting attendees; 2) ministerial portfolios used to code individual ministers; and 3) political parties used to code donor recipients. Figure 1 models the relationships between the fact tables (in orange) and the dimension tables for political contributions data.

Data organisation models for political donations

To build the dimension tables, all values for each relevant variable (for instance a list of every donor) were copied from the fact table into a new table and duplicates were removed (case sensitive). Each unique value in the dimension table was subsequently assigned relevant attributes as detailed below.

For the lobbying fact table, there were 13,731 unique groups of meeting attendees, and for the political donation fact table, there were 5718 unique donors. The dimension tables coded each of these actors to a specific industry group (or ‘other’). Two different coding approaches were trialled for the lobbying and political contribution dimension tables.

For this group, we focused on differentiating between health and commercial organisations. Meeting attendees (ranging from a single attendee to more than ten) were reviewed and coded to either: 1) one of the ‘harmful industries’ (tobacco, alcohol, gambling, food & drink), 2) ‘health advocacy’ or 3) ‘other’ (meetings that did not have at least one actor from these groups). Some meetings included multiple actors representing multiple industry interests. If only one of the four ‘harmful industry’ actors was present, the meeting was coded to that industry. Where multiple ‘harmful industry’ actors were present, we prioritised the industry that best aligned with the meeting purpose. Where that information was not available or insufficient, we prioritised the ‘harmful industry’ with the greatest representation in that meeting.

For the political contribution group, we focused on commercial attributes. All donors were first coded to their harmful industry group (or other). For those entities, we then classified them as either company or trade association and whether they were large (> = 0.5% market share), small (< 0.5% market share) or single establishments (e.g., local hotels) (using data from Euromonitor). Some commercial actors could be coded to multiple industries (e.g., Coca-Cola Amatil is both ultra-processed food and alcohol and Australian Hotels Association is both alcohol and gambling). For all commercial actors, a primary industry designation was established in consultation with colleagues at VicHealth and external experts, though we discuss challenges with this approach in the discussion.

There were 43 unique ministers and 88 unique portfolios. Records of ministers’ start and end dates were accessed from the NSW government website. To facilitate analysis and enable comparisons across portfolios, 16 thematic groups of similar portfolios (e.g., health and mental health) were developed.

As of June 2021, Australia had 49 registered political parties (not including 25 state-affiliated parties). There were two main parties represented in the House of Representatives – the Australian Labor Party and a coalition between the Liberal Party of Australia and the Nationals – with five other parties also represented (Australian Greens, One Nation, Katter's Australia Party, Palmer United Party and Centre Alliance). The 1914 unique donor recipients were coded as one of eight major parties (Liberal, Labor, National, Greens, One Nation, Katter's Australia Party, Palmer United Party or Centre Alliance), ‘Minor Party’ or ‘Other’ (e.g., other parties or political organisations) [47]. Each recipient was also coded to its state branch (e.g., AU-QLD), or if it was a federal branch or unspecified parties they were coded to ‘Australia’. All other recipients not linked to a political party were also coded to ‘Australia’ (e.g., 250 Club Limited).

Google Data Studio was used to explore data visualisation. However, Google Data Studio had limitations in the number of dimension tables that can be linked to a fact table, thus a modified approach was used to organise and analyse the data to fit within the requirements of Google Data Studio. Figure 1 shows the simple model of political donations used in the pilot and an expanded model allowing for more relationships between tables.

Google Data Studio was used to chart and visualise the lobbying and political contribution data (and is the source for figures in the results). The lobbying and political contribution data and dimension tables were uploaded to Google sheets. Data Studio allows you to ‘blend’ two tables using one or more ‘join keys,’ to link the fact and dimension tables. The fact table can then be visualised and filtered by the different variables in the dimension table (for example, comparing the alcohol and gambling industries’ meeting with politicians, or comparing the political contributions from companies versus trade associations). For the lobbying and political contribution datasets, a series of different interactive charts were developed. These were subsequently embedded in a website developed to showcase potential outputs and tested with the advisory group. Examples of these interactive charts are included in the results.