Abstract low-poly geometric background in shades of pale blue and white

Data in the Middle: The common language of research

Census Data and Agriculture: Lessons from IASSIST 2017

The intersection of census data and agriculture emerged as a practical research problem at IASSIST 2017, the conference themed “Data in the Middle: The Common Language of Research.” Held in Lawrence, Kansas, from 23–26 May 2017, it brought together work on big data, deep learning, digital agriculture and global food security. Across plenaries, workshops and parallel sessions, the recurring idea was that evidence becomes more useful when disciplines can interpret one another’s records without erasing their differences.

For an Australian audience, this is not an abstract issue. The Australian Bureau of Statistics (ABS) conducts a population census every five years, while agricultural censuses and farm surveys track land, production and livestock. Linking those sources can reveal how settlement, services, employment and food production interact, but only if privacy, geography, time periods and definitions are handled carefully. The conference offered a useful lesson: interoperability is both a technical design choice and a matter of trust.

Shared questions, distinct data worlds

A population census records people and households at a specific reference date. Agricultural records describe properties, farm businesses, animals, crops and production over a year or a season. Their basic units differ: one source centres on residents and dwellings; the other on operational holdings and production areas. Even a common place name can refer to a census urban area, a cadastral parcel, a statistical region or a management boundary.

The archived program showed how these records become powerful when paired with spatial data, satellite imagery and administrative datasets. Deep-learning methods can classify land cover or detect crop patterns at scale, while census information supplies the social and demographic context that a pixel cannot. In global food-security research, that combination can connect production pressure to food access, labour and household vulnerability rather than treating yields as a stand-alone measure.

Yet integration should not blur categories. A smallholding and a commercial grain operation may share a locality but face different economics; a household farm may have no registered agricultural enterprise. Census respondents may also decline linkage, and farm-level records remain confidential. The conference’s interdisciplinary format therefore mattered as much as its tools: analysts need domain experts who can explain what each measure represents, when it was collected and what omissions mean.

A common language for research data

The theme’s “common language” was most useful as a description of data practice. Researchers need shared metadata, controlled vocabularies, geographic crosswalks, versioned extracts, clear licences and provenance statements that show how each variable was produced. These elements allow an agricultural statistic, a census count and a satellite-derived estimate to travel through an analysis without silently changing meaning.

That practical emphasis appears in the archived research data workshop, which records the conference workshop’s focus on tools and routines for managing research information. Its relevance to census–agriculture projects extends beyond software: folders arranged consistently, analysis scripts retained and documentation updated are what make a linkage reproducible when staff or funding periods change.

In Australia, that discipline is especially valuable across agencies and research teams. An ABS geographic identifier, a state agriculture department boundary and an open-data region may need to be reconciled before analysis. A published data dictionary should state units, suppressions, reference dates and collection methods. Where disclosure controls prevent record-level linkage, aggregate alignment and carefully tested assumptions may be safer than a forced match.

Turning separate sources into shared evidence

Shared evidence begins with a question, not a dataset. A study of rural food access might relate household distance to supermarkets to regional production; a study of labour might compare census employment with seasonal work reported by farms. Each comparison needs a common spatial scale, a defensible time window and uncertainty estimates that travel with the figures. The result should show how much apparent association survives changes in geography or missing data.

Techniques such as record linkage, spatial joins, multilevel modelling and small-area estimation can support this work. They also create new risks: false matches, ecological fallacies and unstable estimates in sparsely populated areas. Conference discussions about big data made the central point visible: more observations do not remove bias when coverage, provenance or labels are weak.

A direct comparison clarifies what each stream contributes:

Dimension Population and housing census Agricultural census and surveys Combined value
Primary unit People, households and dwellings Agricultural holdings, livestock, crops and production Connects demographic capacity with farm activity
Time pattern Snapshot at census night, repeated every five years Reference year or seasonal collection, with annual surveys between censuses Separates structural conditions from seasonal variation
Geography ABS boundaries and statistical areas Cadastral land, property extents, catchments and production regions Supports neighbourhood-to-region analysis without implying identical boundaries
Main strengths Broad social coverage and consistent core questions Direct production detail and land-use information Combines social context with measurable output
Main limits Under-counting, self-response error and restricted small-cell data Sampling error, seasonal gaps and commercial confidentiality Requires uncertainty and suppression rules
Linkage method Stable geography, controlled aggregate variables and approved linkage Property identifiers, farm classifications and spatial concordance Produces defensible local estimates while protecting identities

Methods for seasonal, spatial and sensitive data

Agriculture makes time especially difficult. Australian grain and livestock production responds to rainfall, heat and market cycles, while a census provides a fixed snapshot. Agricultural census years and annual ABS farm surveys can bridge some intervals, but they do not create continuous observations. An Australian study must therefore distinguish structural measures, such as the number of workers or area operated, from outputs that can change sharply after drought or flooding.

For spatial questions, uncertainty is not a footnote. Farm boundaries, sampling locations and census geographies overlap imperfectly, particularly across regional Australia. The stochastic geometry literature offers a relevant body of methods for reasoning about spatial patterns, distances and variability; paired with survey weights and spatial models, it can help researchers test whether an apparent cluster is substantive or an artefact of the boundary.

Deep learning deserves a similarly precise role. Satellite and drone imagery can reveal planting, land cover and spatial change that questionnaires miss, yet model performance may weaken across regions, seasons and farm types. Ground truth from censuses and surveys remains essential. Clear validation, bias checks and interpretable outputs are more defensible than treating a large image archive as automatically representative.

Australian conditions that shape the evidence

Australian conditions shape both the questions and the safeguards. ABS farm classifications distinguish broad industries, while local speakers often use terms such as “property,” “station” and “grazing country” in ways that formal categories may not capture. A cattle station in the Northern Territory, a horticulture business around Coffs Harbour and a grain operation in the Darling Downs each operate within national statistical structures but have different scales, seasonality and service constraints.

Weather compounds those differences. The Australian Bureau of Meteorology, drought monitoring and satellite products provide useful environmental context, yet the same rain event affects catchments unevenly. Linking farm records with rainfall, temperature and vegetation indices can improve analysis, provided geographic resolution and time periods match. It can also connect land use to the realities of communities managing drought, flood recovery or bushfire exposure.

Markets add another layer. Wool auctions, livestock sale yards, dairy processors, exporters and retailers operate as distinct chains, and Australian producers often discuss “the market” in terms of a particular commodity rather than one national system. Census evidence on income, occupation and settlement location can help explain workforce availability or participation in these chains, while farm data can show production concentration and regional dependence.

First Nations communities also require governance beyond technical compliance. Aboriginal and Torres Strait Islander organisations may hold authority over cultural information that cannot be treated as an ordinary attribute. Co-design, benefit sharing and appropriate Indigenous data protocols should accompany any project using mapped or linked information. These relationships are especially important in remote areas where national datasets can otherwise appear complete while service access remains fragmented.

Practical priorities for Australian research teams

The conference’s practical legacy can be translated into a disciplined workflow for Australian projects.

Build a dependable data spine

  • Use a data dictionary that records definitions, units, reference periods and missing-value codes.
  • Map every geographic layer to a published concordance before joining datasets.
  • Preserve raw extracts, scripts, validation checks and version histories together.
  • Attach sampling and uncertainty estimates to every modelled census estimate.
  • Apply disclosure controls before sharing identifiable farm or household records.

Turn linked evidence into useful decisions

  • Co-design questions with farmers, agronomists, councils and service providers.
  • Compare census variables with farm measures only across comparable years and regions.
  • Validate remote-sensing outputs against local ground truth and field observations.
  • Report local results with uncertainty, not as deceptively precise point estimates.
  • Build Indigenous governance and community consent into the data lifecycle.

The IASSIST 2017 archive is valuable because it preserves the programme, schedules, plenaries and workshop discussions as a record of how research data were being managed in 2017. Its central proposition remains practical for Australia: population, land and production records can answer more useful questions when their vocabularies, boundaries and limitations are respected. The resulting evidence is not simply larger; it is more credible and easier for communities, agencies and researchers to use.

At the Conference

What attendees experienced in Lawrence

p1 reed

Plenary Sessions

Keynotes from Daniel Reed on data, technology, and culture, and Jennifer Clarke on digital agriculture and the Midwest Big Data Hub.

platinum icpsr

Workshops

Full-day technical workshops on Tuesday, May 23. Morning sessions ran 9:00–12:00 and afternoon sessions 13:00–16:00. Laptops were required.

gold rockhurst

Social Events

An opening reception at The Oread, a banquet, and an optional post-conference tour of Kansas City including Crescent Moon Winery.

Venue & Accommodations

Where the conference took place

silver ifdo

Kansas Union

University of Kansas campus, Lawrence. Main conference venue with check-in on the 4th and 5th floor lobbies.

platinum ddi

The Oread

1200 Oread Avenue, Lawrence. Hosted the opening reception and offered a room block for attendees.

silver ciser

The Eldridge

701 Massachusetts Street, Lawrence. A partner hotel with a reserved room block for conference guests.

platinum icpsr

Springhill & TownePlace Suites

Marriott properties in Lawrence with room blocks reserved under the "KU IASSIST Conference" name.

Program Highlights

Sessions and activities

Plenary Sessions Workshops Poster Presentations Committee Meetings Opening Reception Banquet Tour Kansas City Pecha Kucha Check-In Local Favorites

Getting Here

Lawrence, Kansas

Kansas Union · University of Kansas
1301 Jayhawk Blvd, Lawrence, KS 66045

Kansas City International Airport (MCI) is approximately 50 minutes by car. Lawrence Transit Routes 10 and 11 served the area ($1 exact change).

Plan Your Stay

Accommodation options that were available

silver ifdo

The Eldridge

701 Massachusetts Street, Lawrence, KS 66044. Room block now closed.

platinum ddi

The Oread

1200 Oread Avenue, Lawrence, KS 66044. Room block now closed.

silver ciser

Springhill Suites

Marriott property. Room block reserved under "KU IASSIST Conference."

gold rockhurst

TownePlace Suites

Marriott property. Room block reserved under "KU IASSIST Conference."