Foodbook 2.0: A mixed method for survey data collection

Abstract

Background: Foodbook 2.0 is a population-based food consumption study that was administered in all provinces and territories in Canada from January 2023 to January 2024. It was the second iteration of the study, the first having been conducted in 2014–2015. This iteration of the study provides updated data on food consumption in Canada. This article describes the study methods and compares respondent demographic information between modes of recruitment.

Methods: The study utilized a dual-method recruitment approach, with 75% of the sample coming from a national list of mailing addresses, and 25% from a national list of telephone numbers. Respondents completed the study online or over the telephone. The study examined food, animal, and water exposures during the seven days preceding the interview, as well as consumer food safety knowledge and practices, recent acute gastrointestinal illness, and demographic characteristics.

Results: There were 21,744 participants in the study, with 73% (15,843) of the sample completing the study online, and 27% (5,901) completing the study over the telephone. The dual-method approach to recruitment was successful in reaching the target sample size. Certain demographic groups had higher response rates for one recruitment method over the other.

Conclusion: The data collected in this study on food, animal and water exposures among people in Canada provide essential population reference information for comparison during enteric illness outbreak investigations. This information will inform risk assessments and public health research, supporting interventions aimed at preventing enteric illness in Canada. Evaluation of the modes of recruitment will help inform future iterations of the Foodbook study.

Introduction

Foodbook 2.0 is a population-level food consumption study that was administered in all provinces and territories (P/Ts) in Canada from January 2023 to January 2024 ^{Footnote 1}. The study collected data on food exposures, water exposures, animal-related exposures, consumer food safety knowledge and practices, recent acute gastrointestinal illness, and demographic factors. The first iteration of the study was conducted in 2014–2015 ^{Footnote 2}. Foodbook 2.0 captured updated data on food exposures and data on new exposures of interest since 2015.

The study was the first nationally representative Canadian food consumption study to collect data over a seven-day period. It has provided valuable data that have been used to inform food safety and public health in Canada ^{Footnote 3Footnote 4Footnote 5Footnote 6Footnote 7Footnote 8Footnote 9Footnote 10Footnote 11Footnote 12Footnote 13Footnote 14Footnote 15Footnote 16Footnote 17Footnote 18}. Foodbook has also been used as a hypothesis generation method during outbreak investigations to help identify the source of illness ^{Footnote 5Footnote 19Footnote 20Footnote 21Footnote 22Footnote 23Footnote 24Footnote 25Footnote 26Footnote 27Footnote 28Footnote 29Footnote 30Footnote 31}.

While the first iteration of the study still provides meaningful data, food consumption patterns change over time, and it is important for outbreak investigators to have accurate population reference data. In addition, new food items have been identified as potential sources of enteric illness since the study's completion in 2015. As a result, Foodbook 2.0 provides updated population reference data for comparison during outbreak investigations. The data will also inform risk assessments and public health research to support public health interventions aimed at preventing enteric illness in Canada.

The original Foodbook study was completed entirely via telephone surveys. Data collection for Foodbook 2.0 employed both telephone and online surveys to reflect changing patterns in survey response in Canada. Because Foodbook 2.0 used different recruitment and survey completion methodology compared to the original study, it is important to evaluate the utility of the mixed methods approach in order to inform the methods of future surveys. This article describes these study methods and compares respondent demographic information between modes of recruitment.

Methods

To ensure the precision of Foodbook 2.0 data, a minimum sample size was calculated using a 95% confidence level and a 5% margin of error, resulting in a total target of 19,968 completed surveys. The minimum sample size was calculated to be 384, based on a 95% confidence interval and a 5% margin of error. The minimum sample size was allotted to each P/T, with the remaining sample assigned proportionally based on population size using Statistics Canada estimates. These targets were designed to provide sufficient power to accurately estimate proportions. This is particularly important for the burden of acute gastrointestinal illness estimates, which are rare in the population.

Foodbook 2.0 was first piloted internally to ensure that the questions were interpreted as intended. The survey was then adjusted based on feedback. Following this, the study underwent an initial pilot test from November 1, 2022 to November 14, 2022, and a second pilot test from December 1, 2022 to December 19, 2022. These pilot tests ensured that skip patterns functioned appropriately, that proper consent was obtained, and that respondents correctly and accurately understood both the questions and the nature of the survey. Based on feedback from interviewers and survey respondents, adjustments were made to improve the survey tool. Adjustments included clarifying the wording of questions, shortening the interviewer script, and removing the consumer food safety module from the telephone version of the survey to reduce survey length and optimize response rates. The finalized version of the survey was launched on January 2, 2023, and the study continued until January 7, 2024. The survey tool is available online ^{Footnote 32}.

The questionnaire asked about food, animal and water exposures during the seven days preceding the interview, aligning with the average incubation period of common enteric pathogens. Food categories included vegetables, fruits, nuts, seeds, meats, seafood, eggs, dairy, other foods, and country foods (only asked to respondents residing in the territories). The questionnaire also asked about food purchasing habits, special diets, water exposures, animal and animal food exposures, food safety knowledge and behaviours (online only), and recent gastrointestinal illness. To avoid respondent fatigue, skip patterns were built into the questionnaire. For example, if a respondent indicated they did not eat any beef in the past seven days, they were not asked any of the subsequent questions about consuming specific beef products.

Sampling technique

The study employed a dual sampling frame approach. Specifically, 75% of the sample was drawn from a national list of mailing addresses, while the remaining 25% came from a national list of telephone numbers. Households were contacted either by mailed letter or by telephone. One participant was randomly selected from each household to participate in the study, based on the age and number of household members (Figure 1).

Figure 1 - Text description

This figure shows the recruitment letter sent to households across Canada. The letter introduces the Foodbook survey and outlines how long it will take to complete, and what it will ask about. It then walks the letter recipient through how to identify which member of the household was randomly selected to complete the survey. The selected member was randomly chosen based on the number of household members and their age order. Once the respondent was selected, they were given information on how to complete the survey and were to go if they have any additional questions.

To enhance youth recruitment, a child was selected to complete the survey in 50% of households with children under 18 years of age. Despite this, youth recruitment remained low (the proportion of youth respondents was compared to the proportion of youths in the Canadian population). To increase youth participation, beginning in the fifth month of data collection in households with children, a child was selected to complete the survey 100% of the time.

The survey was completed independently by the respondent via an online survey tool (Voxco; Montréal, Québec), or through a telephone interview with a trained interviewer. The survey was offered in English, French, and Inuktitut. All data were entered into Voxco and were housed within Canada.

Participants who had traveled outside their P/T of residence overnight within the seven days before completing the survey were excluded from the food, animal, and water modules of the survey, and only completed the food safety knowledge and practices and recent acute gastrointestinal illness modules.

Informed consent

Informed consent was obtained from all participants. If an individual was under the age of 16 (18 in Québec), consent to participate was sought from their parent or legal guardian. Proxy respondents were utilized for children and individuals with medical or activity limitations. Participants were informed of the objective of the study, the time required to complete the study, and that the study was managed and funded by the Public Health Agency of Canada (PHAC). There was no direct benefit to the participants taking part in the study, but they were informed that the study would help improve enteric disease research and outbreak response in Canada. The study was approved by the Health Canada Research Ethics Board (REB 2021-007P), as well as the Newfoundland and Labrador Health Research Ethics Authority (HREB 2021.145), the Nunavut Research Institute (05 005 23R-M), and the Nunatsiavut Government Research Advisory Committee (NGRAC-14558480).

Optimization of response rate

To optimize response rates, the researchers offered clear explanations of the study to respondents, scheduled call-backs at more convenient times, and offered both online and telephone survey options to all respondents.

Measurement and analyses

Results were weighted using design weights to account for the sampling design, including disproportional sampling frames, probability of selection, and non-response. Post-stratification weights were then applied to align the sample's demographic distribution (i.e., age group, sex, and household size) with 2021 Canadian census data. Weighted proportions were computed using Stata 15.1 (StataCorp, Texas Station, Texas).

The survey was considered complete once the respondent reached the end of the survey and submitted it. Due to survey skip patterns, respondents may not have answered all questions. For instance, those who answered 'No' to consuming 'Any beef' skipped the rest of the beef section. During data cleaning, those respondents were marked as 'No' for all subsequent questions about beef consumption. These adjustments were made to accurately reflect the proportion of the population that consumed each item. All survey questions were optional; respondents who left any questions unanswered were still considered to have provided a complete response upon submitting the survey.

A sub-analysis was conducted to compare demographic characteristics between respondents that were recruited by mail versus those recruited by telephone call. The proportions of each demographic characteristic for mail-recruited versus telephone-recruited respondents were analyzed using a Wald chi-square test of equal proportions. For example, the proportion of respondents recruited by mail that fell into the 0–9 age group was compared to the proportion of respondents recruited by phone that fell into the 0–9 age group. The Bonferroni correction was applied to ensure the p-value was a conservative estimate and that the results were not inflated, resulting in a p-value of 0.0011. This was used to minimize Type 1 errors. Additional analyses were performed to compare those that switched methods (i.e., were recruited by mail but responded by telephone or vice versa), to those that responded to the survey using the same stream in which they were recruited.

Results

A total of 145,914 valid landline and mobile telephone numbers were contacted, with 4,965 individuals completing the survey over the telephone, yielding a response rate of 3%. Another 196 individuals contacted by telephone completed the study online.

Mail invitations to participate in the study were sent to 189,801 households. Of those, 15,647 individuals completed an online survey, yielding a response rate of 8%. An additional 936 individuals contacted by mail called the survey company to complete the survey via telephone.

In total, 21,744 individuals participated in the study. Sampling targets were met for all P/Ts except for Nunavut (125/400; 31%) and Northwest Territories (321/402; 80%). Interviews were distributed evenly across all 12 calendar months. Among all respondents, 1,424 (7%) reported traveling outside their P/T of residence within the seven days preceding completion of the survey. Overall, data completeness was much better for phone respondents.

Comparison of the demographic characteristics of respondents recruited by mail versus those recruited by telephone is provided in Table 1. Respondents in the 10–19 age group and those 65 years and older were more often recruited by telephone, whereas those in the 20–64 age group were more often recruited by mail. Those reporting East/Southeast Asian, Middle Eastern, or South Asian ethnicities were more likely to be recruited by mail than by telephone. Alternatively, Indigenous people, White people or those reporting more than one ethnicity were more likely to be recruited by telephone. Individuals with a university certificate, diploma or degree; those earning $120,000 or more; and those living in urban areas were more likely to be recruited by mail. In contrast, individuals with less than a high school diploma or with a high school diploma or its equivalent, were more likely to be recruited by telephone. Finally, respondents with a household income less than $40,000, as well as respondents residing in the Northwest Territories or rural areas were more likely to be recruited by telephone.

Respondents were able to choose their preferred response mode, completing the survey online or by telephone regardless of how they were recruited. Overall, 1,132 respondents switched between modes (Table 2). The demographic characteristics of those who were recruited by mail but chose to complete the survey over the telephone was compared to those that were recruited by mail and completed the survey online. Those that were recruited by telephone but chose to complete the survey online were compared to those that were recruited by telephone and completed the survey over the telephone. Those that were recruited by mail but completed the survey over the telephone were significantly more likely to be older adults aged 65 years or older. This group was also significantly more likely to be female, to reside in New Brunswick, to be White, and to have a high school diploma or its equivalent, or less than a high school education. Respondents that were recruited by mail but completed the survey over the telephone were significantly more likely to be in the 65+ age group.

Discussion

Foodbook 2.0 used a mixed-mode sampling frame designed to collect updated data on food, water, and animal exposures from a representative sample of people in Canada. By combining mail-based and telephone-based recruitment and offering both online and telephone response options, the study aimed to increase participation and reach diverse segments of the population. This approach reflects current best practices in survey methodology and acknowledges the declining efficacy of single-mode surveys ^{Footnote 33Footnote 34}.

The mail-based mode recruited the most participants (73%), particularly adults aged 20–64, and offered a cost-effective mechanism to recruit a large number of households. The mail-recruited, online-completed version of the survey could be completed on the respondent's own schedule, making it more accessible to many. However, mail invitations could not be personally addressed, potentially lowering the response rate. In addition, respondents were instructed to select a household member based on age order; however, the extent to which these instructions were followed is unclear. Data completeness was lower for surveys completed online, as respondents could leave questions unanswered.

In contrast, telephone recruitment was more effective in engaging older adults who may be less comfortable using online platforms, as well as individuals with lower income or educational attainment ^{Footnote 35Footnote 36Footnote 37Footnote 38}. This difference in demographic representation across modes highlights the importance of maintaining multiple recruitment and response channels to ensure inclusivity. Furthermore, offering a telephone version of the survey removed technical barriers (such as those associated with the online survey), making it more accessible to some demographic groups. Having the interviews completed by trained interviewers also ensured the survey was navigated correctly, and that questions were not skipped. However, recruitment and survey completion by phone was extremely time- and resource-intensive, with many calls required to get a single completed survey. Because of the time required to complete the telephone survey, one module of the survey was only asked to online respondents.

Although respondents were able to choose their preferred response mode, switching between modes was uncommon. A larger proportion of mail-recruited respondents chose to complete the survey over the telephone versus telephone-recruited respondents choosing to complete the survey online. Individuals choosing to complete the survey over the telephone were more commonly older adults (65+) or individuals with less formal education ^{Footnote 36Footnote 37Footnote 38Footnote 39}. This ability to switch modes underscores the value of flexible data collection approaches ^{Footnote 33Footnote 34}. Offering both methods may have increased the overall survey response rate, as those that switched methods between recruitment and survey completion may not have completed the survey otherwise. A mixed methods approach ensures the study population is more representative of the Canadian population, which is important for a study like Foodbook, which aims to be nationally representative.

Limitations

Despite the large sample size and national reach, certain limitations remain. Response rates were low, at approximately 8% for mail-recruited online respondents and 3% for telephone recruitment and completion. While the response rate is in line with that of the first iteration of Foodbook (2.7%), this raises concerns about non-response bias ^{Footnote 40}. Individuals who chose not to participate may differ systematically in their exposure patterns, which could impact generalizability. Comparing to census values shows that demographic groups, including older adults, White participants, university-educated individuals, and those from higher-income households were over-represented in the study population. While weighting adjustments were applied to correct for these imbalances, residual biases may persist. Excluded populations, such as those without a fixed home address or telephone access, or those that are not fluent in English, French, or Inuktitut, may also have affected national representation. Furthermore, despite selection methods intended to increase youth participations, youth remained under-represented relative to the Canadian population.

The mixed-mode design introduces potential inconsistencies in how questions were interpreted or answered. For example, online respondents could pause and verify their answers, whereas telephone respondents may have relied on memory alone. Conversely, surveys completed by telephone with an interviewer allowed the interviewer to clarify ambiguous questions and encourage completion, potentially reducing item non-response.

Recruitment in Nunavut and the Northwest Territories was challenging. Given the small population size of the two territories, additional sample and outreach was limited. The sampling plan also had to accommodate restrictions on how many contacts could be recruited in a week, to ensure data collection could be spread throughout the year. These populations are also commonly over-sampled in surveys, which could lead to survey fatigue. Face-to-face interviews and community engagement are recommended approaches for outreach in these areas; however they were not feasible for this study given its national scope ^{Footnote 41Footnote 42}.

Despite these challenges, Foodbook 2.0 represents a significant achievement in large-scale population surveillance. It provides timely, detailed, and geographically comprehensive data on exposures relevant to enteric illness risk in Canada. These data will serve as essential reference points for outbreak investigations, risk assessments, and public health research. The flexibility and scalability of the survey design offer a promising framework for future national health surveillance initiatives.

Future iterations of Foodbook should aim to improve representation of under-sampled populations, including Indigenous communities, recent immigrants, and individuals with limited access to traditional communication channels. Enhanced personalization of recruitment materials, expanded language offerings, and engagement with community organizations may help to address these gaps. Increasing response rates through targeted follow-up strategies and continued innovation in survey delivery will also be important.

Conclusion

Foodbook 2.0 successfully provided updated data on food, water, and animal exposures among people in Canada. It demonstrates the value of flexible, mixed-mode data collection for public health surveillance, particularly in the context of evolving communication preferences and demographic diversity. Although some limitations in reach and representation remain, the survey provides a critical foundation for informing enteric disease prevention and response in Canada.

Authors' statement

MT — Data curation, writing–original draft, writing–review & editing
MR — Data curation, writing–original draft
MKT — Writing–review & editing
AN — Writing–review & editing
VM — Conceptualization, data duration, writing–review & editing

Competing interests

The authors have no conflicts of interest to declare.

ORCID numbers

Megan Tooby — 0000-0001-6154-7155
M Kate Thomas — 0000-0002-3855-0433
Andrea Nesbitt — 0000-0002-4629-8856
Vanessa Morton — 0000-0002-4471-1985

Acknowledgements

The authors would like to acknowledge the important role of the survey respondents who provided the data that are used in this report.

Funding

This work was supported by the Public Health Agency of Canada.

• Previous Page
• Table of contents
• Next page