Copyright is owned by the Author of the thesis.  Permission is given for 
a copy to be downloaded by an individual for the purpose of research and 
private study only.  The thesis may not be reproduced elsewhere without 
the permission of the Author. 
 



  
 

 

 

 

Development and Validation of a Semi-
quantitative Food Frequency Questionnaire 

to Assess Dietary Intake of Adult Women 
Living in New Zealand 

 

A thesis presented in partial fulfillment of the requirements for the degree of  

 

 

Masters of Science  

In  

Nutrition and Dietetics  

 

Massey University, Albany 

New Zealand  

 

 

Zara Leigh Houston 

2014  

  



I 
 

Abstract 
 

Background: There has been an increase in diet-related disease over the last decade 

(University of Otago & Ministry of Health, 2011).  Food frequency questionnaires (FFQ) are 

commonly used to investigate the dietary intake of large populations, due to ease of 

administration and cost effectiveness.  At present in New Zealand (NZ), an up-to-date, 

culturally appropriate food frequency questionnaire (FFQ) for assessing dietary intake is 

lacking.  

Objectives: To develop and validate a culturally appropriate, computerised, semi-quantitative 

food frequency questionnaire to assess the dietary intake of young adult women living in New 

Zealand; to assess the dietary intake of this population using the questionnaire.   

Methods: Participants (n = 110) were women (16 - 45 years) of Māori, Pacific or European 

ethnicity. They completed the New Zealand Women’s Food Frequency Questionnaire 

(NZWFFQ) assessing dietary intake over the previous month, and a four-day weighed food 

record. Validity was evaluated by comparing nutrient intakes from the FFQ with the food 

record using paired t-tests, Pearson’s correlation coefficients, cross-classification, weighted 

kappa and Bland-Altman analysis. Validity was assessed for raw data, and data adjusted to 

account for fruit and vegetable intakes. 

 

Results: Nutrient intakes were significantly higher from the NZWFFQ data compared with the 

food record for all nutrients except monounsaturated fat, polyunsaturated fat and alcohol (p 

< 0.05). Pearson’s correlation coefficients ranged from 0.10 (iron) to 0.80 (vitamin A) with an 

average of 0.39 ± 0.14. Correct quartile classification ranged from 22% (phosphorus) to 47% 

(saturated fat). Correct classification into same and adjacent quartiles ranged from 62% (iron) 

to 86% (saturated fat). Gross misclassification into opposite quartile ranged from 3% 

(saturated fat) to 10% (iron). For weighted Kappa, saturated fat had moderate agreement (ĸ 

= 0.41 - 0.6), and other nutrients had fair agreement (ĸ = 0.21 - 0.4). These findings only 

differed marginally following fruit and vegetable adjustment, with the exception of vitamin A 

in which validity measures decreased.  



II 
 

Conclusion: The NZWFFQ had good relative validity for ranking individuals by dietary intake, 

and was able to categorise participants with higher or lower intake than reference ranges. 

Similarly to previous literature, The NZWFFQ overestimated dietary intake. Therefore, it is not 

suitable for assessing absolute dietary intakes. 



III 
 

Acknowledgements 

There are a number of people I would like to acknowledge for their involvement in this study. 

Firstly, I would like to thank the participants involved in this research, who were part of the 

wider EXPLORE study, without their participation this would not have been possible.  

Thank you to my two academic supervisors, Dr Kathryn Beck and Dr Rozanne Kruger, whose 

extensive knowledge, support and patience were the backbone of this study.   

I would also like to thank Wendy O’Brien and Shakeela Jayasinghe, who coordinated the 

recruitment, screening and testing of participants. Thank you to the EXPLORE team, who 

helped facilitate participant testing; Dr Pam von Hurst, Dr Cathryn Conlon, Richard Swift, 

Owen Mugridge, PC Tong, Maria Casale and Andrea Fenner.  

I would also like to thank AJ Hepburn, Sarah Philipsen, Jenna Schrijvers and Chelsea Symons 

for their help with the endless hours of data entry.  

Lastly, thank you to my family and friends for their endless support and positivity.  

 

 

 
 

 

 



IV 
 

Table of Contents 
Abstract .................................................................................................................................................. I 

Acknowledgements ........................................................................................................................... III 

Table of Contents  ............................................................................................................................... III 

List of Tables  ...................................................................................................................................... III 

List of Figures  ..................................................................................................................................... III 

Abbreviations  ................................................................................................................................... VIII 

 

Chapter 1 Introduction ...................................................................................................................... 1 

1.1. Background ................................................................................................................................ 1 

1.2. Purpose of the Study ................................................................................................................. 3 

1.3. Aim .............................................................................................................................................. 4 

1.4. Thesis Structure ........................................................................................................................ 4 

1.5. Researchers’ Contributions ...................................................................................................... 5 

 

Chapter 2 Literature Review ............................................................................................................. 6 

2.1. Introduction ............................................................................................................................... 6 

2.2. Dietary Intake and Health in Young Women Introduction .................................................... 6 

2.3. Dietary Assessment Methods ................................................................................................... 7 

2.4. Dietary Assessment Challenges in the Adult Population ..................................................... 10 

2.5. Development of a food frequency questionnaire .................................................................. 15 

2.6. Components of a Food Frequency Questionnaire  a FFQ ..................................................... 16 

2.7. Assessing the Validity of an FFQ ............................................................................................ 21 

2.8. International FFQ’s Exploring Dietary Intake ....................................................................... 28 

2.9. FFQ’s Available for Use in Adult Female New Zealanders ................................................... 35 

 

Chapter 3 Methods ............................................................................................................................ 40 

3.1. EXPLORE Study........................................................................................................................ 40 

3.2.       Ethical Approval ...................................................................................................................... 40 

3.3. Development of the New Zealand Women’s Food Frequency Questionnaire .................... 40 

3.4. Pre-testing the Questionniare ................................................................................................ 40 

3.5.       Computerised Format ............................................................................................................. 44 

3.6. Validation of the NZ Women’s Food Frequency Questionnaire ........................................... 45 



V 
 

Chapter 4 Results.................................................................................................................................. 52 

4.1. Participant Characteristics ..................................................................................................... 52 

4.2. Energy, Macro and Micronutrient Intake .............................................................................. 53 

4.3. Validity of the New Zealand Women’s Food Frequency Questionnaire .............................. 57 

4.4. Goldberg Equation for Under-reporters ................................................................................ 64 

 

Chapter 5 Discussion .......................................................................................................................... 65 

5.1. Overall Findings ....................................................................................................................... 65 

5.2. Energy, Macro and Micronutrient Intake .............................................................................. 65 

5.3. Validity of the New Zealand Women’s Food Frequency Questionnaire .............................. 66 

5.4. Adjustment for Fruit and Vegetable Intake ........................................................................... 69 

5.5. Energy Adjustment .................................................................................................................. 70 

5.6. Under-reporting ...................................................................................................................... 71 

 

Chapter 6 Conclusions ...................................................................................................................... 73 

6.1. Study Strengths and Limitations ............................................................................................ 73 

6.2. Recommendations for Future Research ................................................................................ 79 

6.3. Conclusion ................................................................................................................................ 80 

 

References .......................................................................................................................................... 81 

 

Appendices ......................................................................................................................................... 93 

Appendix A. Standard Operating Procedure for the New Zealand Women’s Food Frequency     
Questionniare  ...................................................................................................................................... 93 

Appendix B. New Zealand Women’s Food Frequency Questionnaire  ............................................... 98 

Appendix C. Standard Operating Procedure for the Weighed Four-day Food Record  .................. 164 

Appendix D. Four-day Weighed Food Record  .................................................................................. 165 

 

 

 

 



VI 
 

List of Tables 
Chapter 2 ................................................................................................................................................ 6 

Table 2.1.        Advantages and Disadvantages of Varoius Dietary Assessment Methods ...................... 9 

Table 2.2.        International Food Frequency Questionniare Validation Studies .................................. 31 

Table 2.3.        Food Frequency Questionnaires Available for Use in New Zealand Adult Females ....... 37 

 

Chapter 4 .............................................................................................................................................. 52 

Table 4.1.        Participant Characteristics ............................................................................................. 52 

Table 4.2.        Mean Daily Dietary Intake from the Questionnaire and Food Record vs. New Zealand          
National Recommendations .................................................................................................................. 55 

Table 4.3.        Comparison of Mean Daily Intakes from the Questionnaire and Food Record ............. 58 

Table 4.4.        Pearson’s Correlation Coefficients, Cross-classification and Weighted Kappa for Dietary      
Intake between the Questionnaire and Food Record............................................................................ 61 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



VII 
 

List of Figures 
Chapter 2 ................................................................................................................................................ 6 

Figure 2.1.        Stages to Consider i nthe Development of a Food Frequency Quesitonniare ............... 15 

 

Chapter 3 .............................................................................................................................................. 39 

Figure 3.1.        Study Process for Questionnaire Development and Validation .................................... 40 

 

Chapter 4 .............................................................................................................................................. 51 

Figure 4.1.        Bland-Altman Plots of Relative Validity for Dietary Intake from the Questionnaire and 
Food Record .......................................................................................................................................... 62 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



VIII 
 

Abbreviation List  

4DFR    Four-day Food Record  

AARP    American Association of Retired Persons (Cohort) 

AI    Adequate Intake  

AMDR    Acceptable Macronutrient Distribution Range  

BF%    Body Fat Percentage 

BIA¹    Bioelectrical Impedance Analysis  

BMI    Body Mass Index  

BMR    Basal Metabolic Rate  

BOD POD¹    Air Displacement Plethysmography  

CI    Confidence Interval 

CVD    Cardiovascular Disease  

DEXA¹    Dual Energy X-ray Absorptiometry  

EAR    Estimated Average Requirement  

EXPLORE study  Examining Predictors Linking Obesity Related Elements 

FFQ    Food Frequency Questionnaire  



IX 
 

ĸ     Kappa statistic 

LER    Low Energy Reporter (of dietary intake) 

LOA    Limits of Agreement  

NHMRC   National Health and Medical Research Council 

NRV    Nutrient Reference Value 

NZ    New Zealand 

NZANS    New Zealand Adult Nutrition Survey 

NZEU    New Zealand European 

NZRD    New Zealand Registered Dietitian 

NZWFFQ   New Zealand Women’s Food Frequency Questionnaire  

p    p-value (statistical analysis) 

PAL    Physical Activity Level  

Pr(a)²    Relative observed agreement  

Pr(e)²    Hypothetical probability of chance agreement 

r    Correlation coefficient (statistical analysis) 

RDI    Recommended Daily Intake  



X 
 

RMR    Resting Metabolic Rate 

SD    Standard Deviation 

SOP    Standard Operating Procedure  

Note. ¹Are methods of measuring body composition (BIA, BOD POD and DEXA); ²Are components of 
Goldberg’s cut-off method measuring under-reporters of dietary intake.   



1 
 

1. Introduction  

 

1.1. Background  

With growing evidence of the link between diet, health and disease, it is becoming 

increasingly important to assess dietary intake. For chronic diseases such as obesity, type 2 

diabetes, cardiovascular disease (CVD) and some cancers, diet is a key modifiable risk factor 

(Alwan, 2001). Dietary assessment enables further exploration and understanding of the links 

between dietary intake, health and disease, with the potential to influence health status 

worldwide (Chiuve et al., 2011; Gonzalez, 2006).  

There has been a considerable change in food availability and dietary patterns over recent 

years (University of Otago & Ministry of Health, 2011), with increases in portion sizes, 

snacking and consumption of meals prepared outside the home.  These changes have resulted 

in a greater proportion of energy now being provided by foods that are nutrient-poor and 

energy-dense such as fast foods, sugar-sweetened drinks, cakes, salty snacks, biscuits and 

confectionary. It has been suggested that we now live in an ‘obesegenic environment’ in 

which weight gain is promoted (Giskes et al., 2011; Swinburn, 2008).  

Furthermore, over the last decade there has been a concurrent increase in diet-related 

disease (Ministry of Health, 2013; Russell et al., 1999; University of Otago & Ministry of 

Health, 2011).  Obesity (body mass index (BMI) over 30 kg/m²) (World Health Organisation, 

2007) in adult New Zealand (NZ) women (15+ years) has increased from 21% in 1997, to 32% 

in 2008/2009 (Statistics NZ, 2014). These rates are even higher within some ethnic groups, 

with 73% of Pacific, and 54% of Māori females now obese (Statistics NZ, 2014). Obesity has a 

negative impact on health; increasing the risk of developing type 2 diabetes, cardiovascular 

disease and some cancers (Statistics NZ, 2014). This negative impact is evident through the 

increased prevalence of Type 2 diabetes. Over 13% of Pacific females aged over 15 now have 

Type 2 diabetes; these rates are more than four times higher than those of non-Pacific 

females (Ministry of Health, 2013).  



2 
 

Paradoxically, micronutrient intake remains a problem. It is estimated that between 6 - 34% 

of females aged 15 - 50 years have an inadequate intake of dietary iron. Furthermore, iron 

deficiency (serum ferritin < 12 μg/L and zinc protoporphyrin > 60 μmol/mol) affects between 

10 - 12% of this female age group (Otago University & Ministry of Health, 2011). Inadequate 

intake is even higher for the micronutrient calcium, with 56 - 88% of females aged 15 - 50 

years not meeting national recommendations (Otago University & Ministry of Health, 2011).  

 

In order to further establish associations between dietary intake, health and disease, it is 

important to develop validated dietary assessment tools.  Traditionally, food records have 

been the preferred method of dietary intake assessment. The food record is a prospective 

method of dietary assessment, in which the participant records all foods and beverages 

consumed over a period of one or more days. Details that are recorded include; the types and 

brands of foods consumed, exact food quantities measured at the time of intake, food 

preparation and cooking methods.  However, there is a large participant burden with this 

method, and multiple days of data collection and processing require extensive effort and cost 

to the research team (Willet, 2013).  

Therefore, when investigating the dietary intake of large populations, the food frequency 

questionnaire (FFQ) is now commonly used as it is more cost effective and less time 

consuming for both the participant and research team (Gibson, 2005; Thompson & Subar, 

2008; Willet, 2013; Willet, 1998). Usual intake is better approximated with the FFQ, a 

retrospective method which enables intake to be assessed over a longer period in comparison 

to the food record. It is important to note that the FFQ is not a suitable tool for estimation of 

actual nutrient intake, rather it is able to categorise participants based on their intake (high, 

moderate, low), and also identify those at the upper and lower extremes (Thompson & Subar, 

2008; Willet, 2013; Willet, 1998). 

 

The FFQ is provided in a survey format, often computerised, with a list of foods presented to 

the participant. Participants are required to select how often each food item is consumed 

(e.g. never, twice a month, once a week, once a day). Some FFQs have an additional 

component assessing the portion size of food items. In a quantitative FFQ, the participant is 

required to self-select their portion size, whereas a semi-quantitative FFQ provides a standard 



3 
 

(albeit subjective) portion size. Although more dietary information may be ascertained with 

the inclusion of portion sizes, the estimation involved is one of the greatest sources of error 

in dietary assessment (Willet, 1998).  

Food frequency questionnaires need to be current and specific for the population of interest, 

as food preferences, availability, dietary patterns and intake differ not only between 

individuals, but also over time (Cade et al., 2004; Cade et al., 2002; Willett, 1998). 

Furthermore, FFQs are culturally specific, performing differently across subcultures within a 

population. Mayer-Davis et al. (1999) found lower correlations between the FFQ and the 

reference method (24-hour recalls) in Hispanics and African Americans when compared to 

those found for non-Hispanic white populations, ranging between 0.21 - 0.72 and 0.24 - 0.89 

respectively. A NZ study by Metcalf et al. (1997) also found lower correlations for Māori and 

Pacific subgroups which ranged between 0.36 - 0.63, in comparison to those for NZ Europeans 

between 0.41 - 0.65. These findings highlight the importance of including culturally relevant 

foods in a FFQ (Cade et al., 2002; Willet, 2013). 

 

It is vital that a FFQ is assessed for validity in the population of interest before it is used.  Even 

small, subtle design changes can affect the performance of a FFQ (Cade et al., 2004, 2002). 

Validity involves comparison with another dietary assessment method such as a food record, 

ensuring the FFQ measures what it is intended to measure (Willet, 2013). The comparison can 

be undertaken using a range of statistical methods including comparison of group means, 

correlation coefficients, cross-classification, weighted kappa statistics and Bland-Altman 

analysis. There is no single ‘gold standard’ method of statistical analysis, thus it is 

recommended that dietary validation studies use a combination of statistical methods, rather 

than one in isolation (Willet, 2013).  

1.2. Purpose of the Study  

At present in NZ, an up-to-date, culturally appropriate FFQ is lacking. Although FFQs have 

been developed to evaluate multi-nutrient intake in NZ adults, only five have investigated 

validity (Bell et al., 1999; Bolch, 1994; Metcalf et al., 1997; Sam et al., 2012; Sharpe et al., 

1993). Four of these FFQs are not suitable for use, as validation was undertaken at least 15 

years ago. The only recent FFQ evaluated the diet of a population who were male and female, 



4 
 

aged 30 - 59 years and primarily NZ European (Sam et al., 2012). The total proportion of Māori 

and Pacific participants was 5%, much lower than the total NZ national representation of 

22.3% (Statistics NZ, 2014). Therefore, this FFQ may not be suitable to assess varying nutrient 

intake across cultures in the NZ population. These limitations highlight the need for a 

validated, culturally appropriate FFQ to assess the nutrient intake of young adult women 

living in NZ.  Such an FFQ will be able to be used in future studies, exploring associations 

between nutrient intake, health and disease in young adult women living in NZ.  

1.3. Aim 

To develop and validate a culturally appropriate semi-quantitative food frequency 

questionnaire for use in young adult women living in NZ.  

1.3.1. Objectives  

 To develop a culturally appropriate, computerised, semi-quantitative food frequency 

questionnaire to assess the dietary intake of young adult women living in NZ.  

 To assess the dietary intake of a convenience sample of young adult women living in NZ. 

 To validate the computerised, semi-quantitative food frequency questionnaire (using a 

four day weighed food record) for dietary intake in young adult women living in NZ. 

1.4. Thesis Structure 

This study has been structured into six chapters. Chapter one introduces concepts covered in 

this research and highlights the purpose of the study. The second chapter is a review of the 

literature, covering the relationship between diet and health, dietary assessment methods 

and challenges, development and validation of a food frequency questionnaire (FFQ), and a 

review of previously validated FFQs. The third chapter details and justifies the methodology 

used to develop and validate a FFQ. Chapter four reports the results of this study. This is 

followed by chapter five which discusses the findings from this study. To conclude, chapter 

six is a summary of the study, including a reflection of strengths, limitations and 

recommendations for future use and research.  



5 
 

1.5. Researchers’ Contributions  

Table 1.1: Researchers Contributions to this Study 

Researchers Contributions to the Thesis  
Zara Houston Designed and led the NZWFFQ validation study, developed the 

NZWFFQ recruited participants, screened participants, 
supervised participant testing, data collection of the NZWFFQ 
and 4DFR, data entry of the NZWFFQ and 4DFR, data analysis, 
statistical analysis, interpretation of results. 

Dr Kathryn Beck Academic supervisor and assistance/ guidance of: research 
design, NZWFFQ development, methods  and protocols, 
statistical analysis, results interpretation, thesis revision and 
approval. 

Dr Rozanne Kruger Application for ethics, Academic supervisor and assistance/ 
guidance of: research design, NZWFFQ development, methods  
and protocols, results interpretation, thesis revision and 
approval. Principal investigator for EXPLORE study. 

Adrianna Hepburn and Sarah 
Philipsen 

Assistance with data entry of the 4DFR. 

Jenna Schrijvers and Chelsea 
Symons 

Assistance with data entry of the NZWFFQ. 

 Wider EXPLORE study Contributions  
(note: participants were recruited from this wider study and 
completed the semi-quantitative questionnaire during EXPLORE 
testing) 

EXPLORE co-ordinators: 
Wendy O’Brien and Shakeela 
Jayasinghe 

Coordinated EXPLORE participant recruitment, screening and 
testing.  

EXPLORE recruitment and 
screening: Wendy O’Brien, 
Shakeela Jayasinghe, Zara 
Houston, Sarah Philipsen, 
Adrianna Hepburn, Rozanne 
Kruger 

Participant recruitment and screening across Auckland. 

EXPLORE testing: Wendy 
O’Brien, Shakeela Jayasinghe, 
Zara Houston, Pam von Hurst, 
Cathryn Conlon, Richard Swift, 
Owen Mugridge, Maria Casale, 
Andrea Fenner, Adrianna 
Hepburn, Sarah Philipsen, 
Jenna Schrijvers, Rozanne 
Kruger, Kathryn Beck  

Participant testing which included eight stations measuring: 
blood pressure, taste perception, three measures of body 
composition and three dietary questionnaires (one of which was 
the semi-quantitative questionnaire).  

PC Tong  Assistance with Survey Monkey, equipment for data collection.  
Note. NZWFFQ = New Zealand Women’s Food Frequency Questionnaire; 4DFR = four day weighed 
food record.  



6 
 

2. Literature Review 
 

2.1. Introduction  

This review of literature explores several aspects related to dietary assessment in young 

women.  Within this literature review, associations between dietary intake, health and 

disease outcomes in young women are reviewed, emphasising the importance of accurate 

dietary assessment methodology.  The various methods and challenges of assessing dietary 

intake in individuals and groups are then explored, with a particular focus on the use of food 

frequency questionnaires (FFQs).  This is followed by a review of the factors to consider when 

developing a FFQ, and methods of validation for a newly developed FFQ. 

The following online databases were systematically searched for relevant literature: PubMed, 

Web of Science and Google Scholar. The publication period ranged from 1961 to 2014. The 

search was undertaken in reverse chronological order using the key search terms: food 

frequency questionnaire, diet assessment, valid, female, disease, health, diet trends and New 

Zealand. Key terms were also used in combination with the two functions; ‘AND’ ‘OR’. Full 

text English journal articles that matched the search criteria were reviewed, as were relevant 

citing articles. A manual search was also undertaken using reference lists from recent review 

articles and validation studies to identify further literature.  

2.2. Dietary Intake and Health in Young Women Introduction 

There is a growing body of evidence regarding the relationship between diet, health and 

disease. One of the key determinants of health is nutrition, therefore, it is important to be 

able to assess dietary intake accurately to further explore this relationship in young women.  

Nutrition-related risk factors (such as Body Mass Index (BMI), blood pressure, fruit and 

vegetable intake, and total blood cholesterol) are known to cause diabetes, cancer, stroke, 

heart disease and other major health issues (Ministry of Health, 2013). These dietary risk 

factors collectively account for over 11% of health loss in New Zealand (NZ) i.e. healthy life 

that is lost to illness, premature death or disability (Ministry of Health, 2013; World Health 

Organisation, 2002). Obesity in NZ women has increased by 150% from 21 to 32% over the 

last decade, and diabetes in Pacific women is now at an all-time high of 13% (Statistics NZ, 



7 
 

2014). By assessing dietary intake, there is potential to improve these modifiable risk factors 

in young NZ women, reducing the future burden of disease.   

Paradoxically, micronutrient intake remains a problem. The results of the New Zealand Adult 

Nutrition Survey 2008/09 (NZANS) found the prevalence of iron deficiency (i.e. serum ferritin 

< 12 μg/L, and zinc protoporphyrin > 60 μmol/mol) in NZ females to have doubled from 2.9% 

to 7.2% since 1997 (Otago University & Ministry of Health, 2011). Rates are highest in females 

aged 15 - 18 (10.6%) and 31 - 50 (12.1%) years. These groups also had the lowest dietary 

intake of iron, with an estimated 34.2% and 15.4% respectively, not meeting national daily 

recommendations (Otago University & Ministry of Health, 2011). 

 

Dietary calcium intake is also an issue for NZ females. The NZANS (08/09) found that 73% of 

females had an inadequate intake of calcium. Of particular concern is the sub-group of 15 - 

18 year old Pacific and Māori females, in which 95% had an inadequate intake of calcium 

(Otago University & Ministry of Health, 2011). Young adulthood is the period where peak 

bone mass is developed.  A low calcium intake puts these females at greater risk of 

osteoporosis in later life (Mann & Truswell, 2007; Ministry of Health, 2013).  

2.3. Dietary Assessment Methods 

The four main dietary assessment methods that are commonly used include food records, 24-

hour recall, diet history and the food frequency questionnaire (FFQ). The advantages and 

disadvantages of the four dietary assessment methods are outlined in table 2.1. 

Food records and 24-hour recalls often assess dietary intake over short periods, capturing the 

specific foods an individual has consumed over the duration of one or more days. With a 

weighed food record, the participant is required  to directly weigh and record all of the food 

and beverages they consume, ideally completed at the time of consumption to remove 

reliance on memory. With an estimated food record the participant simply records food and 

beverages in terms of household portion sizes, ideally as they are consumed.  Reported 

intakes may decrease due to fatigue if recording is required longer than four consecutive days 

(Gersovitz et al., 1978; Gibson, 2005; Willet, 2013). Non-consecutive days of reporting can be 

beneficial to reduce participant burden.  



8 
 

The 24-hour recall is used to assess what the participant has consumed during the previous 

24 hours. This method relies on memory, and is conducted in an interview format, usually by 

a trained interviewer (Thompson & Subar, 2008).  

In contrast, the FFQ and diet history explore the individual’s usual food intake over a longer 

period of time (Willet, 2013). The FFQ is used to assess how frequently a list of food items is 

consumed over a given period i.e. a month. It is provided in a survey format and is often self-

administered. The diet history is usually performed by a trained interviewer, who conducts a 

structured interview of questions regarding habitual food intake from the core food groups 

(e.g. fruit and vegetables, bread and cereals etc.) over the past seven days. A cross check is 

then conducted, where further detail is obtained about usual intake over the past three to 12 

months. The cross check may also explore the makeup of meals and cooking techniques 

(Thompson & Subar, 2008).  A diet history is often used by dietitians in the clinical setting.   

In recent years as technology has developed, new dietary assessment methods are emerging 

e.g. the use of digital images and mobile phone applications. Digital images can be taken using 

mobile phones or digital and disposable cameras to capture food selection, meals and plate 

waste. Some validity studies assessing digital images have found energy and nutrient intakes 

to be significantly underestimated in comparison to the reference method (Kikunaga et al., 

2007; Lassen et al., 2010; Martin et al., 2009). Other emerging tools are phone applications 

that function as food records. They use built-in cameras, integrated image analysis and a 

nutrient database that allows users to record foods that have been consumed (Zhu et al., 

2010). At present, these methods are very time consuming for the researcher, but with future 

technological advancement their use may become more common.  

 

 



9 
 Ta

bl
e 

2.
1:

 A
dv

an
ta

ge
s a

nd
 D

isa
dv

an
ta

ge
s o

f V
ar

io
us

 D
ie

ta
ry

 A
ss

es
sm

en
t M

et
ho

ds
 (a

da
pt

ed
 fr

om
 T

ho
m

ps
on

 &
 S

ub
ar

, 2
00

8)
  

Di
et

ar
y 

As
se

ss
m

en
t M

et
ho

ds
 

Ad
va

nt
ag

es
 

Di
sa

dv
an

ta
ge

s 
FF

Q
 

 
To

ta
l d

ie
ta

ry
 in

fo
rm

at
io

n 
ob

ta
in

ed
 o

ve
r t

he
 

pa
st

 1
 m

on
th

 to
 1

 y
ea

r  
 

Lo
w

er
 co

m
pa

ra
tiv

e 
pa

rt
ici

pa
nt

 b
ur

de
n 

 
 

As
se

ss
es

 u
su

al
 in

di
vi

du
al

 d
ie

ta
ry

 in
ta

ke
 

 
Ea

tin
g 

be
ha

vi
ou

r i
s n

ot
 a

ffe
ct

ed
 

 
Lo

w
er

 co
st

 to
 re

se
ar

ch
er

 
 

Lo
w

er
 re

se
ar

ch
er

 a
nd

 p
ar

tic
ip

an
t b

ur
de

n 

 
In

ta
ke

 m
ay

 b
e 

m
isr

ep
or

te
d 

 
 

Pr
ec

isi
on

 is
 n

ot
 q

ua
nt

ifi
ab

le
  

 
Co

gn
iti

ve
ly

 d
iff

icu
lt 

fo
r p

ar
tic

ip
an

ts
 

 
De

pe
nd

en
t o

n 
pa

rt
ici

pa
nt

s’ 
m

em
or

y 
 

Re
se

ar
ch

er
 b

ur
de

n 
fo

r q
ua

lit
y 

co
nt

ro
l o

f d
at

a 
ca

pt
ur

in
g 

an
d 

en
te

rin
g/

 cl
ea

ni
ng

 o
f d

at
a 

Fo
od

 re
co

rd
s 

 
No

t r
el

ia
nt

 o
n 

m
em

or
y 

 
O

fte
n 

us
ed

 in
 v

al
id

at
io

n 
st

ud
ie

s a
s a

 
re

fe
re

nc
e 

m
et

ho
d 

- l
ea

st
 co

rr
el

at
ed

 e
rr

or
s 

w
ith

 o
th

er
 m

et
ho

ds
 d

ep
en

de
nt

 o
n 

m
em

or
y 

e.
g.

 F
FQ

  
 

Q
ua

nt
ifi

ab
le

 d
ie

ta
ry

 in
ta

ke
  

 
Po

te
nt

ia
l t

o 
en

ha
nc

e 
be

ha
vi

ou
r c

ha
ng

e 
or

 
w

ei
gh

t c
on

tr
ol

 th
ro

ug
h 

se
lf-

m
on

ito
rin

g 
 

 
Pa

rt
ici

pa
nt

 a
nd

 re
se

ar
ch

er
 b

ur
de

n 
is 

hi
gh

 
 

Pa
rt

ici
pa

nt
 tr

ai
ni

ng
 re

qu
ire

d,
 m

us
t b

e 
m

ot
iv

at
ed

 a
nd

 li
te

ra
te

 
 

Po
te

nt
ia

l n
on

-re
sp

on
se

 b
ia

s t
hu

s n
on

-re
pr

es
en

ta
tiv

e 
sa

m
pl

e 
 

Co
st

 to
 re

se
ar

ch
er

 h
ig

h 
 

Us
ua

l i
nt

ak
e 

re
qu

ire
s m

an
y 

re
co

rd
 d

ay
s 

 
Re

du
ct

io
n 

of
 in

ta
ke

 re
po

rt
in

g 
w

ith
 ti

m
e 

 
An

 in
cr

ea
se

 in
 d

ay
s r

ec
or

de
d 

in
cr

ea
se

s a
tt

rit
io

n 
 

 
Ea

tin
g 

be
ha

vi
ou

r m
ay

 b
e 

af
fe

ct
ed

 
 

Un
de

r-r
ep

or
tin

g 
of

 in
ta

ke
 is

 co
m

m
on

 
24

 h
ou

r r
ec

al
ls 

 
Ea

tin
g 

be
ha

vi
ou

r n
ot

 a
ffe

ct
ed

 
 

Lo
w

 p
ar

tic
ip

an
t b

ur
de

n 
 

 
Q

ua
nt

ifi
ab

le
 d

ie
ta

ry
 in

ta
ke

  
 

Re
du

ce
d 

no
n-

re
sp

on
se

 b
ia

s, 
as

 p
er

fo
rm

ed
 

by
 in

te
rv

ie
w

er
/ r

es
ea

rc
he

r  
 

 
Ap

pr
op

ria
te

 a
cr

os
s p

op
ul

at
io

ns
 

 
Hi

gh
 co

st
 to

 re
se

ar
ch

er
 

 
Un

de
r-r

ep
or

tin
g 

of
 d

ie
ta

ry
 in

ta
ke

  
 

Us
ua

l i
nt

ak
e 

re
qu

ire
s m

an
y 

re
ca

ll 
da

ys
  

 
De

pe
nd

en
t o

n 
pa

rt
ici

pa
nt

s m
em

or
y 

 

Di
et

 h
ist

or
y 

 
To

ta
l d

ie
ta

ry
 in

fo
rm

at
io

n 
ob

ta
in

ed
  o

ve
r t

he
 

pa
st

 1
 m

on
th

 to
 1

 y
ea

r  
 

Fo
od

 co
ns

um
pt

io
n 

in
fo

rm
at

io
n 

pe
r m

ea
l 

 
As

se
ss

es
 u

su
al

 in
di

vi
du

al
 d

ie
ta

ry
 in

ta
ke

 
 

Ea
tin

g 
be

ha
vi

ou
r i

s n
ot

 a
ffe

ct
ed

 

 
Hi

gh
 b

ur
de

n 
to

 re
se

ar
ch

er
 

 
De

pe
nd

en
t o

n 
pa

rt
ici

pa
nt

s’ 
m

em
or

y 
 

O
fte

n 
in

ta
ke

 is
 m

isr
ep

or
te

d 
 

 
Pr

ec
isi

on
 is

 n
ot

 q
ua

nt
ifi

ab
le

  
 

Co
gn

iti
ve

ly
 d

iff
icu

lt 
fo

r p
ar

tic
ip

an
t 



10 
 

2.4. Dietary Assessment Challenges in the Adult Population  

There are a number of challenges associated with assessing the dietary intake of adult 

populations.  These include under-reporting, dietary variation, quantifying supplement use, 

participant and researcher error, analysis of dietary intake and selecting a dietary assessment 

method. These challenges are reviewed in the following sections.  

2.4.1. Under-reporting  

Under-reporting (or mis-reporting) is a common problem with dietary surveys. In Western 

populations there has been an increase in the degree of under-reporting over the last two-

to-three decades (Heitmann & Lissner, 1995; Hirvonen et al., 1997; Ministry of Health, 2013), 

with some studies describing levels of under-reporting as high as 25% in some participants. 

Furthermore, 77% of women were identified as under-reporters in the study by Rennie et al. 

(2007). This increase is believed to reflect not only the increased obesity prevalence, but also 

increased awareness of what constitutes a healthy diet i.e. social desirability bias (Ministry of 

Health, 2003).  

 Current national data on dietary intake is provided by the New Zealand Adult Nutrition 

Surveys (NZANS). These surveys are undertaken every 10 years, using either FFQs or 24-hour 

dietary recalls (Otago University & Ministry of Health, 2011).  Under-reporting in the most 

recent NZANS was investigated by Gemming et al. (2014). The study found reported energy 

intake had decreased from 1997 to 2008/09 however, body weight and Body Mass Index 

(BMI) of the NZ population markedly increased over this time (Statistics NZ, 2014). Analysis 

identified respondents that were low energy reporters (LERs) i.e. participants whose ratio of 

energy intake to estimated Resting Metabolic Rate (RMR) was less than 0.9. More women 

(25%) than men (21%) from the NZANS 08/09 were classified as LERs. The prevalence of LERs 

was significantly greater in people who were either obese (BMI ≥ 30 kg/m² (30%) or 

overweight (BMI 25 - 29.99 kg/m²) (25%), in comparison to those with a normal BMI (18.5 -

24.99 kg/m²) (16%) (p < 0.001). These findings are supported by other studies in women living 

in developed countries (Olafsdottir et al., 2006; Kretsch et al 1999).  There was also a 

significantly greater LER prevalence in Pacific people (33%) in comparison to Māori (26%) (p 

< 0.01), and NZEU (23%) (p < 0.001). Furthermore, there was a substantial increase in LER 



       
 

11 
 

prevalence from the NZANS 97/98 to the NZANS 08/09 across most population groups 

(Gemming et al., 2014). These findings highlight the need for studies to take under-reporting 

into account when assessing dietary intake in populations.  

2.4.2. Dietary Variation  

Dietary intake varies from day to day in free-living individuals, over and above an underlying 

consistent dietary pattern (Willet, 2013). If daily intake was random, with no element of 

consistency, then it would not be possible to measure nutrient effects epidemiologically. 

Dietary variation can be influenced by systematic factors such as season and days of the week 

e.g. fruit and vegetable intake can vary depending on the season (Ziegler et al., 1984), and 

intake can differ between week days and the weekend (Willet, 2013). Some studies suggest 

variations in total energy intake also occur during the menstrual cycle (Davidsen et al., 2007) 

however, this influence is minor in comparison to other dietary intake factors such as physical 

activity (Willet, 2013). 

Variation in day-to-day intake also differs across nutrients, and this needs to be considered 

when assessing dietary intake (Cade et al., 2002). For example, total energy varies the least 

due to well-regulated physiological mechanisms. Macronutrients vary moderately day-to-

day, and contribute largely to energy intake. In contrast, micronutrients have a large variation 

as they are often concentrated in particular foods.  Daily intakes of these nutrients can be 

high or low depending on an individuals’ food selections, requiring multiple days of dietary 

assessment to determine ‘usual intake’ across nutrients (Willet, 2013). Willet recommends at 

least 28 days for assessment of vitamin A intake due to seasonal differences in intake, and 

concentrations in specific foods e.g. liver that are not eaten daily. Although long periods of 

dietary assessment are ideal, this is often not feasible in epidemiologic research due to the 

increased burden on participants, and the duration required for researchers to process the 

data (Cade et al., 2002). Therefore, samples of one or a few days are usually measured. Short 

sampling methods influence the distribution or spread of intakes across a population, 

increasing the standard deviation. This effect is demonstrated with participants on the 

extremes; those with low true intakes will have days when intake is lower than their long-

term average, and those with high true intakes will have days when intake is above their long-



       
 

12 
 

term average. When only a short duration of intake is assessed, the likelihood of this effect is 

greater, thus, the distribution of intakes across the population will increase (Willet, 2013).   

2.4.3. Quantifying Supplement Use 

Dietary supplements are becoming increasingly popular, with thousands of commercial 

products now readily available in pharmacy’s, supplement stores, supermarkets and online. 

The nutrients in these products come in varying quantities and different chemical forms 

(Zerwekh, 2008). Therefore, it is difficult to quantify how much supplements contribute to 

an individual’s measured dietary intake. Although consideration of supplements is 

recommended when assessing dietary intake (Roswall et al., 2010), the FFQ validation 

studies included in this literature review did not investigate supplement use.  

 

2.4.4. Participant and Researcher Error   

Systematic and random errors occur in all methods of dietary assessment. These can include 

errors in recording, estimation of portion sizes and using national food composition tables. 

The quality of data collected depends on the processes used, ensuring the assessment 

method is suitable for the population, having trained staff, and using standardised protocols 

i.e. standard operating procedures (SOPs) (Willet, 2013).  

2.4.5. Analysis of Dietary Intake 

A common error that applies to both food records and FFQs is the food composition database 

which is used for dietary analysis. Ideally, the same composition database should be used for 

both dietary methods, as this will match errors, and decrease the effect of such errors on the 

validity assessment. Selection of an appropriate database should consider the following: using 

a national database, relevant population-specific foods, and appropriate nutrients for local 

foods. This database should be regularly updated as the composition of foods change 

frequently through continuous development of new foods, and changes in food preparation 

methods (Margetts & Nelson, 2010). Between databases there will be variations in what foods 

have been included, and also the nutrient content of foods due to differences in analysis 

methods (Greenfield & Southgate, 2003). Thus, the importance of the nutrients’ accuracy and 

completeness should be considered when selecting the database (Thompson & Buyers, 1994). 



       
 

13 
 

Accuracy will be reduced if too many foods need to be substituted because their absent from 

the database (Greenfield & Southgate, 2003). 

2.4.6. Selecting a Dietary Assessment Method 

The choice of which dietary assessment method to use is largely dependent upon the purpose 

and objectives the study. Time and budgetary considerations are often factors, with FFQ’s 

typically imposing less researcher and participant burden compared to food records. 

Characteristics of the target population should be considered (Willet, 2013).  For example, 

the pilot investigation from the Nurses’ Health study revealed that after undertaking an FFQ, 

only 40% of participants would complete a one week food record (Stryker et al., 1991). 

Furthermore, a food record was only provided by 14% of participants from the American 

Association of Retired Persons (AARP) cohort who took part in a small pilot investigation 

(Willett, 2008). Willet (2013) believes the perceived burden for participants to complete a 

food record or 24-hour diet recall is exacerbated by the rigorous procedures they involve. It 

is thought that motivation may be enhanced by the introduction of an engaging or fun 

element to electronic versions, such as creating avatars (Willet, 2013).  

When investigating large populations in epidemiological studies, the FFQ is now commonly 

used as it is cost effective to process and distribute in comparison to other more intensive 

dietary assessment methods (Gibson, 2005; Thompson & Subar, 2008; Willet, 1998). Willet 

(2013) states that “Since the 1980s, substantial refinement, modification, and evaluation of 

food frequency questionnaires have occurred, so that data derived from their use have 

become considerably more interpretable” (Willet, 2013, p. 71).  

2.4.6.1. The Food Frequency Questionnaire 

The FFQ is provided in a survey format, where a list of foods is presented to the participant. 

Participants are required to select how often each food item is consumed e.g. never, twice a 

month, once a week, once a day etc. The food list should be specific to the purpose of the 

study, and to the population of interest (Thompson & Subar, 2008).  

Traditionally, the descriptive qualitative design of the FFQ provided information on food 

consumption patterns. These simple frequency questionnaires assessed how often a given list 

of foods was consumed. Developments were then made by adding portion sizes to provide 



       
 

14 
 

nutrient information, encompassing a quantitative design. Quantitative FFQs are discussed in 

further detail in section 2.6.3. ‘Portion Size Estimation’. 

Supplementary questions may be included in the FFQ. These often cover preparation 

methods, cooking methods, food types and brands. Cross-check questions can also be 

included, which assess how frequently whole food groups are consumed e.g. ‘how many 

serves of vegetables do you usually eat each week’. The agreement between this cross-check 

question and individual vegetable item questions can then be assessed. Some FFQs provide 

an open-ended section, allowing participants to report foods and frequencies of items not 

included on the food list. This ensures that the participant’s total diet is captured, and it may 

identify individuals who consume foods that are not included in the FFQ (Calvert et al., 1997).  

Average long-term or usual dietary intake is better approximated with an FFQ, which assesses 

a longer period of diet exposure compared with other dietary assessment methods that only 

assess a few specific days (e.g. food record or 24-hour recall). The trade-off is that crude 

information is gathered over an extended duration in exchange for precise information over 

a very short duration. A description of the usual frequency that a food is consumed relies on 

generic memory, and is believed to be easier for participants as opposed to describing specific 

foods that were consumed during a past meal, which relies on episodic memory. Cognitive 

research supports this concept (Bradburn et al., 1987; Smith, 1993; Smith et al., 1991). This 

method also captures nutrients where intake varies from day-to-day such as iron.  Such 

nutrients are difficult to measure with the 24-hour recall and food record (Willet, 2013).  

Literate participants can complete the FFQ on paper or online in a self-administered format.  

Alternatively, the FFQ can be administered by an interviewer, although this is associated with 

cost and time implications. The FFQ can be difficult to administer for populations with low 

literacy. It has been suggested by Cade et al (2002) that in these populations, a 24-hour recall 

may be a more suitable dietary assessment method. 

 

As outlined in table 2.1 above, there are weaknesses in all methods of dietary assessment. 

The FFQ is a retrospective method, dependent upon the participant’s memory and accuracy 

in estimating portion sizes. Therefore, the food categories should be clear and familiar to the 



       
 

15 
 

population of interest (Sam et al., 2012), and it is also crucial to include ethnically traditional 

foods which contribute significantly to nutrient intake (Willet, 2013).  

 

2.5. Development of a food frequency questionnaire 

There are multiple stages involved in the development of any dietary assessment method. 

Initially a literature review is undertaken; the stages that then follow depend on the method 

selected and the purpose of the investigation. These are outlined in figure 2.1.  

 

Stages to Consider in the Development of a Food Frequency Questionnaire 
 

 

Figure 2.1: Process involved in the development of an FFQ.  

 

 

 

Literature review on dietary assessment, methodolgy, 
validation and the population of interest 

Questionnaire development; including food groups, food 
list, quantities/ portion sizes

Questionnaire pilot test - final modifications made and  
review by research team

Questionnaire administration; computerised format

FFQ validation with a reference method e.g. weighed food 
record or biomarker



       
 

16 
 

2.6. Components of a Food Frequency Questionnaire  a FFQ 

The questionnaire design is dependent upon the objectives of the study i.e. whether intake 

of specific foods will be measured (single-nutrient FFQ) or conversely, whether a more 

comprehensive dietary intake assessment will be undertaken (multi-nutrient FFQ). Multi-

nutrient FFQs are usually the most desirable option due to difficulties anticipating which 

dietary questions will be important upon analysis. For example, some important food items 

may be excluded in overly restrictive food lists (Willet, 2013).  

Another questionnaire design consideration is whether ranking individuals dietary intake is 

the primary objective, or alternatively, measuring absolute intake. Ranking of intake is often 

used in epidemiological research, and enables categorisation of participants with higher or 

lower intake than reference ranges, and identification of those at the extreme ends of 

nutrient intake distribution.  Ranking is performed using statistical methods, for example 

correlation coefficients (r) can be used to express the accuracy of ranking between the dietary 

tool (e.g. the FFQ), and the reference method (e.g. food record). The r values can then be 

interpreted in relation to the proportion of participants who are correctly classified into the 

bottom or top quartile of the distribution of energy or nutrient intakes. For studies comparing 

health characteristics and diet within groups, correct classification is important (Nelson et al., 

1989). In contrast, absolute intake compares the actual intake obtained from the dietary tool 

with the reference method. Statistical methods can be used to assess the agreement between 

two dietary methods, and include mean nutrient data and Bland-Altman statistics.  

Once the objectives of the FFQ design have been established, the development of the FFQ 

involves three main components; the food list, frequency-response options and portion size 

estimation (Willet, 2013). The following sections explore factors that need to be considered 

within each of these three components.  

 

2.6.1. The Food List 

The food list is the first step in developing an FFQ.  It is important to consider food groups, 

the number of foods items within each group and the order of such food items.  



       
 

17 
 

It is critical that the food list is assembled in an unambiguous and clear format. The FFQ should 

aim to capture the required information from a minimal amount of questions (Cade et al., 

2004), with careful selection of the most informative food items (Willet, 2013). Willet (2013) 

has suggested that there are three general characteristics of an ‘informative’ food: that it is 

used often by a number of participants, that it has a substantial nutrient content for that 

being investigated, and that variations in use differ across participants. Some examples of 

informative foods include bread varieties, cheese or bananas. Participants may fatigue with a 

FFQ which contains too many food items; losing interest and or focus, leading to mistakes 

(University of Otago & Ministry of Health, 2011). The food list is designed to capture both the 

key nutrients of interest, and commonly eaten foods of the population. Cade et al. (2002) 

performed a review of over 200 FFQs and found that FFQ food lists had a median of 79 items, 

and varied from between 5 to 350 items. The main determinant of the questionnaire length 

was the number of nutrients that the FFQs were designed to assess e.g. single versus multi-

nutrient. It was previously suggested by Willet et al. (1998) that 100 food items represented 

the cut-off point at which the quality of answers would reduce thereafter due to boredom 

and fatigue. Willet (2013) now believes the limit may be approached at 130 food items, and 

that participants who are willing to undertake long dietary questionnaires often have a strong 

interest in nutrition.  

It is also important to have an ordered and structured food list, as the interpretation of a food 

item can be influenced by another. Related food items need to be clustered together in same 

food category (Willet, 2013). For example, within these food categories, specific foods e.g. 

chocolate muesli bar should appear before general foods e.g. the standard muesli bar (Cade 

et al., 2004). 

The most important food groups should be found near the start however, they also should 

not be the very first questions. This is due to the high question error rate initially, until 

participants understand how to respond.  Further errors rise again at the end of the 

questionnaire when fatigue sets in (de Castro, 1991). Comprehension and interpretation is 

enhanced when single foods are questioned in a simple and clear manner, in comparison to 

long complex questions that include multiple foods (Andersen et al., 2002; Bascand, 2011; 

Cena et al., 2008; Peterson et al., 2008; Thompson et al., 2008, 1994; Thurnham, 1988; Willet, 

2013). For example, if the researcher wanted to know how often plums, peaches and apricots 



       
 

18 
 

were consumed it would be easier to ask the participants three simple questions on each food 

item, rather than having to consider the fruits separately, integrating all three frequencies 

and provide an average frequency (Willet, 2013).  Whilst having three separate questions in 

this example does make the FFQ longer, it may in fact be quicker for the participant to 

complete due to ease of comprehension. 

Following these considerations, the food list can be compiled using several approaches. The 

food list from a pre-existing FFQ can be modified to include or exclude food items relevant to 

the population of interest. Alternatively, published food composition tables can be used to 

identify foods containing the nutrients of interest. However, this method will include foods 

which have a high concentration of some nutrients e.g. iron in black pudding, but that are not 

consumed frequently enough to contribute to the data. It has been suggested that pilot 

testing the questionnaire in a similar population (to that of interest) can help this issue, as 

infrequently consumed foods can be deleted, or a stepwise regression analysis can be 

undertaken to identify discriminating foods (Heady, 1961). 

2.6.2. Frequency Response Options 

The next step in development of the FFQ is the frequency response options. These can either 

be given in a closed or open-ended format. The closed format often uses multiple choice 

questions asking participants how often they consume the item with given responses e.g. 

ranging from never through to multiple times daily. This reduces coding time, and increases 

the completion rate. Whilst open-ended questions have the potential to obtain more 

information, they often have lower completion rates, more transcription errors, and increase 

the coding time. Due to this, the preferable option is multiple choice responses (Bascand, 

2011; Cade et al., 2004; Thurnham, 1988). 

Throughout the literature, multiple choice questions usually offer between five to ten 

frequency options. The FFQ must capture inter-subject variation; including foods frequently 

consumed e.g. bread, through to foods rarely eaten with high nutrient contents e.g. liver. Due 

to this, the range of frequency options generally start from ‘never’ and progress to ‘six plus 

times daily’. However, it has been suggested that non-beverage items are sufficiently 

captured with a maximum frequency option of ‘two plus times daily’ (Bascand, 2011; Cade et 

al., 2004). Nutrient intake is primarily provided by commonly consumed foods, therefore, it 



       
 

19 
 

is important to ensure higher weekly frequencies are emphasised, and that gaps are not 

present between the frequency options (Thurnham, 1988; University of Otago & Ministry of 

Health, 2011).  

Multiple choice questions can be presented in either a horizontal or vertical format. The 

horizontal layout provides the frequency options beside the food item. In contrast, the 

vertical layout provides a vertical list of frequency options below each food item. It is easier 

for the elderly, children and populations of lower education to complete these questions as a 

vertical list (Willet, 2013).  

2.6.3. Portion Size Estimation  

The final step of FFQ development is the allocation of a portion size to food items. This 

depends on whether the FFQ is a simple, semi-quantitative or quantitative FFQ. The simple 

FFQ does not collect any additional portion size information, participants are simply required 

to select how frequently each food item is consumed (Willet, 2013).  

A semi-quantitative FFQ uses a given portion size, and requires the participant to report how 

often the given portion is consumed (Cade et al., 2002). The total question number, and 

therefore length of the questionnaire can be reduced through the incorporation of portion 

sizes into frequency response questions. This does not necessarily reduce the duration of the 

questionnaire however, as some participants may find the two-part (combination) question 

confusing, particularly when the given portion size differs from what they consume. The semi-

quantitative FFQ assumes that the participant will account for differing portion sizes by 

adjusting their frequency response. In reality, this does not always occur, and numerous 

participants disregard the given portion size (Cade et al., 2004).  

Additional clarity can be provided from the question when the food items are those with 

natural units, for example one egg, glass of milk, cup of coffee. Willet (2013) suggest questions 

should be posed in this way for such food items, as additional information is provided without 

having to add another question for unit size specification. Using a ‘palm size’ or ‘palm volume’ 

to estimate portion size may be easier for participants to visualise, particularly in NZ where 

nutritional food magazines such as the Healthy Food Guide often use this as a portion guide. 

However, the validity of this measure is yet to be proven.  



       
 

20 
 

A quantitative FFQ uses either open-ended portion size questions or multiple choice portion 

options e.g. small, medium, and large (Block et al., 1986; Hankin et al., 1983). However, 

participant interpretation of these portion sizes is subjective, and may differ significantly 

between participants (Willet et al., 1985). Open-ended questions may include measurement 

aids that are provided in relation to the portion size description (Cade et al., 2002) e.g. 

photographs, household measures such as measuring cups and spoons or lifelike food models 

(Morgan et al., 1978). Open-ended questions are not often used however, due to the expense 

of coding and processing the responses (Willet, 2013). 

2.6.3.1. Errors in Estimation of Portion Size 

It is the quantification of portion size that produces the largest error of all (Gibson, 2005; 

Young & Nestle, 1995). Issues often arise for food items that do not have natural or typical 

units i.e. pasta or meat. When a portion is given for these items, e.g. 100 g meat, the 

participant is expected to adjust their reported frequency of consumption if it differs from 

that provided. Therefore, someone that usually consumes twice this quantity should double 

their reported frequency. Thompson & Subar (2008) informally evaluated this semi-

quantitative approach in the Nurses’ Health Study Cohort. Although they found consistent, 

correct interpretation for the food items with natural units e.g. an apple, participants often 

ignored the specified portion size for the food items without natural units e.g. grapes.    

The literature suggests that use of measurement aids such as food photographs, result in 

fewer errors in comparison to unaided responses. Furthermore, when multiple portion sizes 

are displayed, even fewer errors occur (Baranowski et al., 2010; Contento, 2008; Katsouyanni 

et al., 1997; Longnecker et al., 1993; Subar et al., 2010; University of Otago & Ministry of 

Health, 2011). Willet (2013) found that when multiple food images (up to eight) were 

presented to participants on a computer in successively larger portions, accuracy was 

markedly improved.  

Alternatively, ‘quantity ranges’ can be used to define the portion size. For example, in a study 

by Subar et al. (1995) participants were required to select their usual portion size from ranges 

of categories e.g. rice: less than 1/2 cup, 1/2 - 1 cup or more than 1 cup. They found that 

providing a range for the portion size improved clarity for participants in comparison to 

providing a medium size reference and having to select small, medium or large. However, for 



       
 

21 
 

self-administered questionnaires, they found that portion size questions in both formats were 

frequently ignored.  

In order to obtain useful portion size information, participants need to be able to relate their 

own dietary habits to the questionnaire at hand, conceptualising portion sizes clearly (Willet, 

2013). This is difficult to achieve, as demonstrated by Guthrie (1984) who found that 

participants generally could not describe their usual portion size. Upon cessation of a meal, 

participants described portion sizes of the foods consumed; accuracy to within 25% was 

achieved by less than half the participants.  

Studies have found little difference in dietary intake results between different methods used 

to assess portion sizes within food frequency questionnaires (Willet, 2013). In a study by 

Samet et al. (1984), correlations for a simple frequency questionnaire versus a quantitative 

format using food models were 0.86 and 0.91 respectively. These findings indicate that both 

dietary tools performed well, with the quantitative tool only marginally better than the simple 

questionnaire. Similar findings were observed in the large German Cohort by Nothlings et al. 

(2007), suggesting that minimal information is obtained from additional portion size 

questions. In contrast, some authors suggest that the largest inter-subject variation arises 

from portion size, thus disputing the value of gathering any data on portion size (Ajani et al., 

1994; Molag et al., 2007). However, in theory greater accuracy is provided when more 

information is obtained. The inclusion of portion size and hence, accuracy improvement, 

could potentially be masked by errors from portion estimation. This leads to the suggestion 

that portion sizes do not need to be removed from FFQs, but further investigation is required 

into methods to improve the accuracy for participant’s estimates of their food portions. This 

may involve measurement aids, which have shown some improvement in accuracy in several 

studies (Baranowski et al, 2010; Contento, 2008; Katsouyanni et al, 1997; Longnecker et al, 

1993; Subar et al, 2010; University of Otago & Ministry of Health, 2011;). There is however, 

no single recommendation on the best measurement aid and further research is required. 

2.7. Assessing the Validity of an FFQ 

“Validity refers to the degree to which the questionnaire actually measures the aspect of diet 

that it was designed to measure. This implies that a comparison is made with a superior, 

although always imperfect, standard” (Willet, 2013, p. 96). 



       
 

22 
 

Assessing dietary intake in epidemiological studies requires a quick, easily applied method 

that is able to obtain accurate information. Validation studies are undertaken to determine 

the method’s accuracy, by comparing the test method to that of a reference method. Validity 

of the dietary method is reflected in the degree of closeness from the test method and 

reference method. This relationship depicts ‘relative validity’ as the reference method is not 

absolutely valid itself. Therefore, relative validity can ascertain whether the FFQ’s answers 

compare to that from the reference method (Cade et al., 2004) 

It is crucial to validate an FFQ before use in the target population, ensuring it measures what 

it is intended to measure. False associations may occur between disease markers and dietary 

intake if the information is incorrect (Cade et al., 2002). Investigations into diet and disease 

associations have the potential to influence health status worldwide; these can be examined 

at the level of dietary patterns, food groups, individual foods and nutrients. Therefore, it is 

important that the FFQ is validated at the level for which it was intended (Willet, 2013). 

There are several factors to consider in the design of a validation study to ensure that the 

data collected represents the results that will be obtained when the FFQ is used in the 

population of interest (Cade et al., 2002). These factors are described in the following sections 

and include the following: study population, sample size, reference methods, recording days 

required, sequence of administration and the statistical analysis.  

 

2.7.1. Study Population 

Validation of the FFQ needs to be undertaken in a sample which represents the population 

that it has been intended for use in. The performance of an FFQ can be affected by even small, 

subtle design changes (Willet, 2013). Due to this specificity, the FFQ may not be appropriate 

for use in differing populations. Therefore, new questionnaires should be validated, as should 

previously validated questionnaires when the target population differs (Willet, 2013). 

Question responses are influenced by age, gender, health status, and demographics (Cade et 

al., 2002). Foods within the FFQ should be culturally specific, as the FFQ may perform 

differently across subcultures within a population (Willet, 2013). The levels of food 

consumption reported in the FFQ may be influenced by differing food perceptions between 



       
 

23 
 

cultures i.e. healthy versus unhealthy foods (Margetts & Nelson, 2010). Furthermore, 

ethnicity and socioeconomic status influence diversity in the diet and hence, affect validity. A 

highly diverse diet decreases agreement, as the opportunities for mistakes are greater when 

compared to a less diverse diet.  

2.7.2. Sample Size 

The requirement for sample size is influenced by precision requirements, statistical methods 

assessing validity, daily nutrient variation, number of recorded days from the reference 

method and the total duration of these recorded days (Thompson & Buyers, 1994; Willet, 

1998). Furthermore, across different populations the variation both between- and within-

subjects differs. Accurately determining the required sample size is therefore difficult.  A 

minimum of 50 participants is required if Bland-Altman statistics are to be used, with the 

suggestion that studies preferably use over 100 (Cade et al., 2002). For correlation 

coefficients, it has been suggested that 150 participants are required as it is only at this point 

that the confidence interval (CI) precision stops increasing (Willet, 1998). However, such 

sample sizes are based on the presumption that food intake has been recorded for more than 

12 days (Cade et al., 2002; Willet, 1998). For most validation studies this is not feasible. There 

is a general consensus that the number of participants required does not need to exceed 100-

200 participants, ensuring that the amount of recorded days is sufficient to describe the 

participant’s usual diet accurately (Cade et al., 2002; Henriquez-Sanchez et al., 2009; Serra-

Majem et al., 2009a; Serra-Majem et al., 2009b; Willet, 1998).  

 

2.7.3. Reference Methods 

There is no dietary assessment method that enables usual dietary intake to be measured with 

absolute accuracy (Cade et al., 2004). A reference method is used for comparison with the 

FFQ to assess its validity. When selecting the reference method, it is crucial that the error 

sources are independent of those found with a FFQ (Cade et al., 2002). The weighed food 

record has the least correlated errors with the FFQ therefore, this dietary tool is most 

commonly used as the reference method used in validation studies (Margetts & Nelson, 

2010). A food record involves weighing and recording all foods and drinks participants 



       
 

24 
 

consume at that given time. Thus, with the food record, errors with estimation of portion size, 

restricted food lists and memory are limited (Cade et al., 2002).  With food records however, 

issues can arise when the participant is unable to weigh their food, relying on estimation of 

portion size. Participant burden is high with this method, therefore, validation studies often 

keep recording days to a minimum (Cade et al., 2002; Willet, 2013). This is discussed in further 

detail in the following section 2.7.4. ‘Recording days required’. 

 

The use of biomarkers is one approach which is independent of dietary intake and thus, not 

associated with FFQ errors. The key limitation of using biomarkers is their nutrient specificity, 

assessment can only be undertaken with some nutrients, as the biomarker must reflect 

dietary intake over the same period as dietary assessment method (Gibson, 2005). For 

example, nutrient levels of vitamin C, vitamin B6 and fat-soluble vitamins are found in plasma 

or serum. Plasma and serum reflect recent intake therefore, the biomarker can only be used 

to validate a short period e.g. that covered by a 24-hour recall or a short food record (Gibson, 

2005). Furthermore, errors arise with biomarkers from food metabolism, influencing the 

relationship with usual dietary intake (Cade et al., 2002). This method is invasive, costly, and 

increases the burden on the participant. Rather than use on its own as a reference method, 

biomarkers are more suitable for use in conjunction with another method.   

In populations where literacy is limited, the dietary recall method can be useful (Cade et al., 

2002). When the participant is unaware that the recall will be undertaken, the likelihood of 

the dietary method influencing the participant’s actual diet is reduced. However, dietary 

recalls depend on estimation of portion size and memory, similar to the FFQ. Due to this, they 

are a less suitable reference method for the FFQ (Cade et al., 2002).  

 

In a review of over 200 FFQ validation studies, Cade et al. (2002) found that food records were 

the most frequently used validation method (51%). However, only half of these food records 

were weighed. The 24-hour recall was used by 22% of studies; the diet history by 6%, and 12% 

used another FFQ. The majority of these validation studies (64%) used only one reference 

method. Only 3% of studies used two methods, both the 24-hour recall and weighed food 

record.  

 



       
 

25 
 

In general, FFQ errors arise from the dependence on estimation of the portion size, food 

consumption recall, and food lists which are incomplete. Due to this, the most suitable 

comparative method is the weighed food record, which is unlikely to have errors which 

correlate with the FFQ (Cade et al., 2002; Willet, 2013). 

2.7.4. Recording Days Required 

The variance ratio, e.g. the intra- to inter-participant variation, influences how many days are 

required for estimation of usual dietary intake (Livingstone et al., 2004).  The dietary intake is 

likely to vary less over a month than it would over a year, as the availability of some foods e.g. 

fruit and vegetables depend on the season. Stram et al. (1995) compiled data from two 

studies to calculate the ‘ideal’ duration of dietary recording for validation studies. They found 

that across most settings, having more than four to five food record days was rarely required. 

Willet (1998) also found that five represents an ideal number of reporting days. Participant 

burden increases above five days, and collection of dietary data becomes less accurate (Stram 

et al., 1995; Willet, 1998). A collection period of more than five days results in lower 

participant completion rates, increased awareness of consumed foods, affecting FFQ 

responses, and an increased likelihood of alterations in the foods consumed (Willet, 1998). 

However, it is important to note that the number of recording days required differs between 

nutrients. For example, nutrients such as vitamin A are found concentrated in specific foods, 

and can be influenced by the season. Studies have found that higher correlations for vitamin 

A are found when at least four weeks of dietary intake data are collected (Willet, 2013).  

 

It has been suggested that the period in which the FFQ covers i.e. one month, should also 

cover the same period of days recorded by the food record i.e. the recorded days spread over 

a month. Furthermore, these days of food recording should not be consecutive, with an even 

spread over the month and across various days of the week. Regarding usual diet, the true 

variability for each participant is better provided by non-consecutive days as dietary intake 

on consecutive days is shown to be correlated. For example, people often eat similar foods 

on consecutive days due to the groceries they have in the house, or leftovers from meals 

consumed the following day (Hartman et al, 1990; Thompson & Buyers, 1994).   



       
 

26 
 

2.7.5. Sequence of Administration  

The order of the FFQ and reference method administration is important, as they may 

influence one another. The period over which the methods are administered is important as 

the participant’s awareness of food consumption may be increased with the effort of the food 

record to weigh all food and drinks consumed. This in turn could increase the FFQ accuracy if 

it was undertaken afterwards. In contrast, if the FFQ is administered prior to the food record, 

it would relate to a different period of dietary assessment with a possible low association of 

dietary intake between the two methods. To reduce the disadvantages from this, the FFQ 

could be administered both prior and post completion of the reference method (Willet, 1998). 

The average of the result can be used, or alternatively, a random selection of either the first 

or second FFQ for each participant could be compared with the reference method. However, 

this is not always feasible for validation studies, due to cost and time implications.  

2.7.6. Statistical Analysis  

The validity of an FFQ can be assessed using a variety of statistical techniques. However, there 

is no consensus on the most appropriate technique to use (Masson et al., 2003).  Therefore a 

combination of approaches is commonly used. These methods are discussed in the following 

sections.  

2.7.6.1. Correlation Coefficients  

For validation of dietary assessment methods, the most common statistical technique used 

are correlation coefficients (Altman & Bland, 1983). The data distribution will determine 

whether Pearson’s’ or Spearman rank correlations are used (normal or non-normal data 

respectively). Spearman correlations enable participants to be ranked, an approach which 

reduces sensitivity to extreme values (Masson et al., 2003). The correlation will reduce 

between the two dietary assessment methods when a large intra-participant variation in 

nutrient intakes exists. From an FFQ review, Cade et al. (2004) have found that when nutrient 

correlations exceed 0.4, validity is accepted. Furthermore, Cohen (1988) and Hopkins et al. 

(2009) have described the following descriptors for correlation coefficients: 0.9 - 1 almost 

perfect; 0.7 - 0.9 very high; 0.5 - 0.7 high; 0.3 - 0.5 moderate; 0.1 - 0.3 low and 0 - 0.1 

insubstantial. There is controversy however, regarding the use of correlations to measure 

validity. It has been argued by Bland & Altman (1986) that a positive correlation will always 



       
 

27 
 

exist when the same thing is measured by the two dietary assessment methods. Furthermore, 

they believe that only associations can be shown by correlations, not agreement. A high 

correlation can still have poor agreement (Altman & Bland, 2002). It is only when very similar 

results are produced that high agreement will exist (Bland & Altman, 1986). In addition, 

correlations are affected by the range of sample values i.e. a large sample will result in a lower 

correlation (Cade et al., 2002; MacIntyre et al., 2007). Due to these factors, some believe 

validity assessment is too flawed using correlation coefficients (Bland & Altman, 1999; Bland 

& Altman, 1986; Cade et al., 2002). They may be more useful when used alongside other 

statistical techniques (Cade et al, 2002; Masson et al, 2003). To date however, correlation 

coefficients have been used in the majority of validation studies, which enables comparisons 

across studies (Cade et al, 2004).  

2.7.6.2. Paired t-test (Comparison of Means) 

The relative validity can be assessed at a group level by comparing medians or means for 

nutrients obtained from the FFQ and food record (Gibson, 2005). The data distribution will 

determine whether the paired t-test or Wilcoxon’s signed rank test is used for this (normal or 

non-normal data respectively) (Lee, 1980). It is important to highlight that a group mean 

comparison such as this, will not provide information at an individual level on the 

questionnaire’s quality, nor the capability of the FFQ to describe the distribution of dietary 

intake (Block & Hartman, 1989). 

2.7.6.3. Cross-classification and Weighted Kappa Statistic 

Cross-classification enables classification of the participant’s nutrient intake into categories 

e.g. quintiles, quartiles or tertiles based on the two dietary assessment methods (Gibson, 

2005). The term ‘correctly classified’ encompasses the participants who had the same 

category classification by the two dietary methods, given as percentage value. The term 

‘grossly misclassified’, also given as a percentage value, encompasses those with opposite 

categories. However, within the percentage agreement, participants will be included when 

their classification was based on chance alone (Willet, 1998). To account for both the correctly 

classified percentage, and the expected participant proportion classified by chance, the Kappa 

statistic is used (Cohen, 1968). The weighted Kappa is suggested for use when assessing 



       
 

28 
 

ordinal variables, as larger differences are given greater emphasis. The Kappa statistic is 

dependent on category weightings, and the category number used. Due to this, limitations 

arise when the category number is high, as the potential for disagreement increases. This 

results in a lower Kappa value, and thus, understates the agreement (Sim & Wright, 2005). 

Across studies the category number and weightings will differ, therefore, comparisons can be 

limited.  

2.7.6.4. Bland-Altman Analysis 

For studies involving method comparison, Bland-Altman (1999) has suggested using the limits 

of agreement. This technique requires plotting the difference of the two dietary methods for 

each nutrient against the average from the two dietary methods. The limits of agreement are 

then calculated with their 95% confidence interval. Outliers are easily observed on the plot, 

as are trends when intake is increased. It has been suggested that with increased nutrient 

intake, a greater difference may be observed between the two methods, in contrast to that 

from participants with smaller nutrient intakes (Cade et al., 2002). This results in the FFQ 

appearing less reliable, with participants consuming a higher nutrient intake in comparison to 

those with lower intakes. In support of Bland & Altman (1986), Cade et al. (2004) also 

recommend that the agreement between dietary assessment methods should be considered 

using this technique.  

2.8. International FFQ’s Exploring Dietary Intake  

Worldwide there has been a considerable change in food availability and dietary patterns in 

recent years (Swinburn, 2008). Furthermore, nutrition related issues such as obesity are also 

a global problem, with the highest rates of obesity found in the U.S, Mexico and New Zealand 

(Statistics NZ, 2014).  A large amount of epidemiological research investigating the 

relationship between diet and disease is being undertaken internationally. The number of FFQ 

validation studies undertaken in NZ is limited. Hence, it is advantageous to review findings 

from international studies on dietary assessment. A small collection of validated FFQs have 

been reviewed in table 2.2. FFQs of primary interest were those with similar demographics to 

the FFQ validation study at present e.g. young female, no chronic disease nor pregnant, and 

a range of ethnicities.  



       
 

29 
 

Across the validity studies, there was a large range of nutrient correlations from comparisons 

of the FFQ with the reference method. Although the majority of studies validated their FFQ 

against a food record, some studies used diet recalls. 

Validity correlations for macronutrients ranged from weak to very high across the nutrients 

(Cohen, 1988; Hopkins et al., 2009). Protein, polyunsaturated fat, carbohydrate and 

cholesterol were commonly found in studies to have the lowest correlations; between -0.04 

to 0.29 (Boucher et al., 2005; Brunner et al., 2005; Friis et al., 1997; Kumanyika et al., 2003; 

Mayer-Davis et al., 1999).  Stronger correlations (over 0.5) were found for fat, saturated fat 

and sucrose (Bingham et al., 1997; Friis et al., 1997; George et al., 2004). 

Validity correlations for micronutrients ranged from weak to strong (Cohen, 1988; Hopkins et 

al., 2009). Vitamin E, A, C, β-carotene, thiamine, folate and sodium were commonly found in 

studies to have the lowest correlations; between -0.04 to 0.27 (Masson et al., 2003; Boucher 

et al., 2005; Brunner et al., 2005; Friis et al., 1997; George et al., 2004; Kumanyika et al., 2003; 

Mayer-Davis et al., 1999). Stronger correlations (over 0.5) were found for calcium, folate and 

vitamin A, C and E (Bingham et al., 1997; Boucher et al., 2005; George et al., 2004; Masson et 

al., 2003; Mayer-Davis et al., 1999).  

George et al. (2004) found that the FFQ overestimated energy intake by 5.5% in the validation 

study, and 3.4% in the cross-validation study when compared to food records (Cross-

validation involved the use of two reference methods in comparison to one method used in 

the validation study). The two studies had different populations, the validation study included 

college students with a lower average BMI than the cross-validation participants who were 

post-partum, low income mothers. The differences in FFQ energy overestimation may be due 

to under-reporting in the higher BMI cross-validation group. Under-reporting is a common 

issue with dietary assessment, particularly participants with an overweight or obese BMI 

(Cade et al., 2002). In further support of differences between BMI sub-populations, 

Kumanyika et al. (2003) found lower energy correlation coefficients (representing possible 

under-reporting) in a group of black Africans with a BMI over 27 kg/m².  

Dietary tools also perform differently across ethnicities, as observed in the study by Mayer-

Davis et al. (1999). They found lower correlations between the FFQ and reference method 

(24-hour recalls) in Hispanics and African Americans, in contrast to those found for white 



       
 

30 
 

Hispanics. These findings are supported by Cade et al. (2002) and Willet (2013) who have 

stated that FFQs are culturally specific, performing differently across subcultures within a 

population. Validity is often lower in subcultures due to a lack of culturally specific food items 

in the FFQ food list.  

Differences within a population were further observed when participants were stratified by 

education level. Mayer-Davis et al. (1999) found the median correlation coefficient across 

nutrients was 0.49. However, within the subpopulation of participants with less than 12 years 

of education, the median correlation coefficient was much lower at 0.25. Kumanyika et al. 

(2003) also reported lower correlation coefficients for calcium, β-carotene and fibre in 

participants with less than 16 years education compared with participants with more than 16 

years education.  

 

 

 

 



31
 

 Ta
bl

e 
2.

2:
 In

te
rn

at
io

na
l F

oo
d 

Fr
eq

ue
nc

y 
Q

ue
st

io
nn

ai
re

 V
al

id
at

io
n 

St
ud

ie
s 

Re
fe

re
nc

e 
Po

pu
la

tio
n 

Ch
ar

ac
te

ris
tic

s  
FF

Q
 D

es
ig

n 
St

ud
y 

De
sig

n 
Fi

nd
in

gs
 

Bo
uc

he
r e

t 
al

. (
20

05
) 

96
F,

 2
5-

74
y,

 g
en

er
al

 
po

pu
la

tio
n,

 O
nt

ar
io

, 
Ca

na
da

 

-1
26

 it
em

 F
FQ

 
m

od
ifi

ed
 fr

om
 B

lo
ck

’s 
fu

ll-
di

et
 F

FQ
 

-C
an

ad
ia

n 
re

le
va

nt
 

fo
od

s  
-9

 fr
eq

ue
nc

y 
op

tio
ns

 &
 

4 
st

an
da

rd
ise

d 
po

rt
io

n 
siz

es
 p

ro
vi

de
d 

 
-S

el
f-a

dm
in

ist
er

ed
 

O
ve

r 6
m

 p
er

io
d 

 
-F

FQ
  

-1
4d

 in
te

rv
al

 - 
1s

t d
ie

t r
ec

al
l (

1 
w

ee
k 

da
y 

+ 
1 

w
ee

ke
nd

) 
-1

8d
 in

te
rv

al
 - 

2n
d 

di
et

 re
ca

ll 
(1

 
w

ee
k 

da
y 

+ 
1 

w
ee

ke
nd

) 

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.0

7 
(c

ho
le

st
er

ol
) -

 0
.4

1 
(c

ar
bo

hy
dr

at
e)

 
-M

icr
on

ut
rie

nt
 C

C 
ra

ng
e:

 0
.2

7 
(β

-c
ar

ot
en

e)
 - 

0.
63

 (f
ol

at
e)

 
 -D

ea
tt

en
ua

te
d 

va
lid

ity
 C

C:
 m

od
er

at
e 

to
 h

ig
h 

– 
m

ed
ia

n 
of

 
0.

59
   

Ge
or

ge
 e

t a
l. 

(2
00

4)
 

-9
5F

, m
ea

n 
of

 2
0y

 
-S

ou
th

w
es

te
rn

 U
S 

co
lle

ge
 st

ud
en

ts
 

-M
ea

n 
BM

I 2
2k

g/
m

² 
 Cr

os
s -

va
lid

at
io

n 
(u

sin
g 

tw
o 

re
fe

re
nc

e 
m

et
ho

ds
)  

-5
0F

 p
os

tp
ar

tu
m

, 
lo

w
-in

co
m

e,
 m

ea
n 

of
 

23
y  

-M
ea

n 
BM

I 
28

.3
kg

/m
² 

-1
95

 
ite

m
 

se
m

i- 
qu

an
tit

at
iv

e 
FF

Q
  

-F
oo

d 
lis

t 
m

od
ifi

ca
tio

n 
of

 H
ea

lth
 H

ab
its

 a
nd

 
Hi

st
or

y 
Q

ue
st

io
nn

ai
re

 
(H

HH
Q

)  
-9

 fr
eq

ue
nc

y 
op

tio
ns

  
-4

 p
or

tio
n 

siz
e 

op
tio

ns
; 

sm
al

l, 
m

ed
iu

m
, l

ar
ge

, 
ex

tr
a -

la
rg

e 

O
ve

r 6
m

 p
er

io
d 

-F
FQ

  
-1

w
 in

te
rv

al
 - 

3-
da

y 
fo

od
 re

co
rd

 
(2

 w
ee

k 
da

ys
 +

 1
 w

ee
ke

nd
)  

 Cr
os

s -
 v

al
id

at
io

n:
  

-3
m

 p
os

tp
ar

tu
m

 - 
1s

t d
ie

t r
ec

al
l 

&
 2

-d
ay

 fo
od

 re
co

rd
  

-6
m

 p
os

tp
ar

tu
m

 –
 F

FQ
, 2

nd
 d

ie
t 

re
ca

ll 
&

 2
-d

ay
 fo

od
 re

co
rd

 

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.3

4 
(c

ho
le

st
er

ol
) –

 0
.5

3 
(S

FA
) 

-M
icr

on
ut

rie
nt

 C
C 

ra
ng

e:
 0

.2
4 

(s
od

iu
m

) -
 0

.6
5 

(v
it 

A)
 

-C
or

re
ct

 cl
as

sif
ica

tio
n:

 7
6%

 
-F

FQ
 e

ne
rg

y 
ov

er
es

tim
at

io
n 

(v
s. 

fo
od

 re
co

rd
) =

 5
.5

%
 

 Cr
os

s V
al

id
at

io
n:

 
-M

ac
ro

nu
tr

ie
nt

 C
C 

ra
ng

e:
 0

.3
6 

(M
UF

A)
 –

 0
.4

8 
(S

FA
) 

-M
icr

on
ut

rie
nt

 C
C 

ra
ng

e:
 0

.4
5,

 ra
ng

e 
of

 0
.2

8 
(s

od
iu

m
) t

o 
0.

59
 (v

it 
E)

 
-C

or
re

ct
 cl

as
sif

ica
tio

n:
 7

9%
 

-F
FQ

 e
ne

rg
y 

ov
er

es
tim

at
io

n 
(v

s. 
fo

od
 re

co
rd

) =
 3

.4
%

 

M
as

so
n 

et
 

al
. (

20
03

) 
-4

0F
, 1

9-
58

y 
-M

ea
n 

BM
I 

25
.1

kg
/m

² 

-1
50

 
ite

m
 

se
m

i- 
qu

an
tit

at
iv

e 
Sc

ot
tis

h 
Co

lla
bo

ra
tiv

e 
Gr

ou
p 

FF
Q

 

-F
FQ

 co
m

pl
et

ed
  

-4
-d

ay
 fo

od
 re

co
rd

  (
3 

w
ee

k 
da

ys
 

+ 
1 

w
ee

ke
nd

 d
ay

)  

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.3

9 
(c

ho
le

st
er

ol
) –

 0
.7

1 
(S

FA
) 

-M
irc

on
ut

rie
nt

 C
C 

ra
ng

e:
 -0

.0
4 

(th
ia

m
in

e)
 –

 0
.7

5 
(c

al
ciu

m
) 

 -C
or

re
ct

 cl
as

sif
ica

tio
n:

 3
5 

(c
ho

le
st

er
ol

) –
 7

8%
 

(m
ag

ne
siu

m
) 



   
   

 
 

32
  

Re
fe

re
nc

e 
Po

pu
la

tio
n 

Ch
ar

ac
te

ris
tic

s  
FF

Q
 D

es
ig

n 
St

ud
y 

De
sig

n 
Fi

nd
in

gs
 

-P
ar

tic
ip

an
ts

 fr
om

 
Al

be
rd

ee
n,

 S
co

tla
nd

 
-F

FQ
 

us
ed

 
co

lo
ur

ed
 

ph
ot

og
ra

ph
s 

of
 

fo
od

 
po

rt
io

ns
 

-C
om

pl
et

ed
 a

t h
om

e 
 

-9
 fr

eq
ue

nc
y 

op
tio

ns
  

In
ta

ke
 o

ve
r 2

-3
m

 

Co
m

pl
et

ed
 e

ith
er

 9
 d

ay
s b

ef
or

e 
or

 9
 d

ay
s a

fte
r t

he
 F

FQ
 

-G
ro

ss
 m

isc
la

ss
ifi

ca
tio

n:
 0

 (a
lco

ho
l &

 N
SP

) -
 3

0%
 

(th
ia

m
in

e)
 

-W
ei

gh
te

d 
ka

pp
a:

 - 
0.

08
 (t

hi
am

in
e)

 –
 0

.6
6 

(m
ag

ne
siu

m
) 

Ku
m

an
yi

ka
 

et
 a

l. 
(2

00
3)

 
-2

45
F,

 2
1-

69
y 

-B
la

ck
 A

fri
ca

ns
, U

S 
 

-F
ou

r s
ub

gr
ou

ps
 b

y 
BM

I &
 a

ge
: 

≥ 
27

 k
g/

m
², 

< 
27

 
kg

/m
²,  

30
-4

1y
, 4

2+
 y

 
-S

tr
at

ifi
ca

tio
n 

by
 

re
sid

en
ce

 (W
es

t, 
M

id
w

es
t, 

No
rt

he
as

t, 
So

ut
h)

  
 -S

tr
at

ifi
ca

tio
n 

by
 

ed
uc

at
io

n 
(≥

 1
6 

ye
ar

s 
or

 <
 1

6 
ye

ar
s)

 

-6
8 

ite
m

 F
FQ

  
-F

oo
d 

lis
t 

m
od

ifi
ca

tio
n 

of
 

Bl
oc

ks
 

Na
tio

na
l 

Ca
nc

er
 I

ns
tit

ut
e 

(N
CI

) 
FF

Q
 

- 
BW

HS
 

(B
la

ck
 

W
om

en
’s 

He
al

th
 S

tu
dy

)  
- 9

 fr
eq

ue
nc

y 
op

tio
ns

  
-P

or
tio

n 
siz

e 
se

le
ct

io
n;

  
sm

al
l, 

m
ed

iu
m

, l
ar

ge
  

- S
el

f-a
dm

in
ist

er
ed

 

-O
ve

r 1
2m

 
-F

FQ
  

-3
m

 in
te

rv
al

 (s
ea

so
na

l) 
– 

1s
t &

 
2n

d 
24

h-
di

et
 re

ca
ll 

 
-3

m
 in

te
rv

al
 –

 1
x 3

-d
ay

 fo
od

 
re

co
rd

 (2
 w

ee
kd

ay
s +

 1
 

w
ee

ke
nd

)  
-3

m
 in

te
rv

al
 –

 3
rd

 2
4h

-d
ie

t r
ec

al
l 

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.2

0 
(p

ro
te

in
 &

 ca
rb

oh
yd

ra
te

) –
 

0.
23

 (S
FA

)  
-M

icr
on

ut
rie

nt
 C

C 
ra

ng
e:

 0
.0

6 
(v

it 
E)

 –
 0

.2
3 

(β
-c

ar
ot

en
e)

 
 St

ra
tif

ica
tio

n:
 

-E
ne

rg
y 

CC
: l

ow
er

 fo
r B

M
I ≥

 2
7 

vs
. B

M
I <

 2
7 

kg
/m

²  
 (r

 =
 

0.
09

 v
s. 

0.
29

) 
-F

ib
re

 C
C:

 lo
w

er
 fo

r <
16

 y
ea

rs
 e

du
ca

tio
n 

gr
ou

p 
-C

al
ciu

m
 &

 β
-c

ar
ot

en
e 

(E
AD

) C
C:

  l
ow

er
 fo

r <
 1

6y
 

ed
uc

at
io

n 
vs

. 1
6+

 a
s f

ol
lo

w
s (

r =
 0

.0
1 

vs
. 0

.3
7 

&
 0

.2
3 

vs
. 

0.
53

 re
sp

ec
tiv

el
y)

 

Br
un

ne
r e

t 
al

. (
20

01
) 

-4
03

F,
 3

9-
61

y 
-P

ar
tic

ip
an

ts
 fr

om
 th

e 
W

hi
te

ha
ll 

II 
lo

ng
itu

di
na

l s
tu

dy
, 

Br
iti

sh
 ci

vi
l s

er
va

nt
s 

-U
K 

-1
27

 
ite

m
 

se
m

i- 
qu

an
tit

at
iv

e 
FF

Q
  

-F
oo

d 
lis

t m
od

ifi
ca

tio
ns

 
to

 
th

e 
US

 
Nu

rs
es

’ 
He

al
th

 S
tu

dy
 F

FQ
 

-C
om

pl
et

ed
 a

t h
om

e 
 

-9
 fr

eq
ue

nc
y 

op
tio

ns
 

-O
ve

r 1
 y

ea
r  

 
-F

FQ
 co

m
pl

et
ed

 +
 7

-d
ay

 fo
od

 
re

co
rd

   

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.2

9 
(p

ro
te

in
) –

 0
.5

6 
(S

FA
)  

-M
icr

on
ut

rie
nt

 C
C 

ra
ng

e:
 0

.2
2 

(fo
la

te
) –

 0
.4

1 
(v

it 
C)

 
 (a

lco
ho

l 0
.8

5)
 

-E
ne

rg
y 

ad
ju

st
ed

 C
C:

 0
.3

3 
(v

it 
E)

 - 
0.

58
 (S

FA
) (

al
co

ho
l 0

.8
3)

 
 -C

or
re

ct
 cl

as
sif

ica
tio

n:
 3

2 
(P

UF
A)

 –
 4

1%
 (S

FA
) 

 (a
lco

ho
l 6

4%
)  

-G
ro

ss
 m

isc
la

ss
ifi

ca
tio

n:
 0

 (a
lco

ho
l) 

- 8
%

 (v
it 

E 
&

 p
ro

te
in

) 



   
   

 
 

33
  

Re
fe

re
nc

e 
Po

pu
la

tio
n 

Ch
ar

ac
te

ris
tic

s  
FF

Q
 D

es
ig

n 
St

ud
y 

De
sig

n 
Fi

nd
in

gs
 

M
ay

er
-D

av
is 

et
 a

l. 
(1

99
9)

  
-1

86
F 

-A
fri

ca
n 

Am
er

ica
n 

O
ak

la
nd

, H
isp

an
ic 

Co
lo

ra
do

, n
on

-
Hi

sp
an

ic 
w

hi
te

 
Co

lo
ra

do
 &

 O
ak

la
nd

, 
US

A  
-In

su
lin

-R
es

ist
an

ce
 

At
he

ro
sc

le
ro

sis
 S

tu
dy

 
(IR

AS
) 

-1
14

 i
te

m
 F

FQ
, 

se
m

i- 
qu

an
tit

at
iv

e 
m

od
ifi

ca
tio

ns
 

to
 

NC
I-

HH
HQ

  
-9

 fr
eq

ue
nc

y 
op

tio
ns

 
-P

or
tio

n 
siz

e 
se

le
ct

io
n;

  
sm

al
l, 

m
ed

iu
m

, l
ar

ge
 

-O
ve

r 1
 y

ea
r  

-8
x 2

4h
-d

ie
t r

ec
al

ls 
co

m
pl

et
ed

 
ev

er
y 

6 
w

ee
ks

 o
ve

r 1
y 

 
(D

ay
 o

f w
ee

k 
ra

nd
om

ly
 se

le
ct

ed
 - 

co
m

pu
te

r g
en

er
at

ed
)  

-T
he

n 
FF

Q
 co

m
pl

et
ed

 

W
hi

te
 - 

O
ak

la
nd

 
-M

ac
ro

nu
tr

ie
nt

 C
C 

ra
ng

e:
 0

.3
 (P

UF
A)

 –
 0

.7
7 

(S
FA

)  
-M

icr
on

ut
rie

nt
 C

C 
ra

ng
e:

 0
.4

 (v
it 

A 
&

E)
 –

 0
.4

8 
(v

it 
C)

 
Af

ric
an

-A
m

er
ica

n 
- O

ak
la

nd
 

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.3

8 
(c

ar
bo

hy
dr

at
e)

 –
 0

.6
2 

(S
FA

)  
-M

icr
on

ut
rie

nt
 C

C 
ra

ng
e:

 0
.2

2 
(v

it 
A&

 C
) –

 0
.6

7 
(v

it 
E)

 
W

hi
te

 - 
Co

lo
ra

do
 

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.2

9 
(P

UF
A)

 –
 0

.6
3 

(S
FA

)  
-M

icr
on

ut
rie

nt
 C

C 
ra

ng
e:

 0
.2

4 
(v

it 
A)

 –
 0

.8
9 

(v
it 

E)
 

Hi
sp

an
ic 

– 
Co

lo
ra

do
  

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.2

1 
(P

UF
A)

 –
 0

.4
4 

(o
le

ic 
ac

id
)  

- M
icr

on
ut

rie
nt

 C
C 

ra
ng

e:
 0

.2
1 

(v
it 

E)
 –

 0
.7

2 
(v

it 
C)

 
-L

ow
er

 C
C 

fo
r <

12
y 

ed
uc

at
io

n 
vs

. >
12

y 
(C

C 
= 

0.
25

 v
s. 

0.
30

) 

Fr
iis

 e
t a

l. 
(1

99
7)

 
-1

22
F,

 2
0-

29
y 

-G
en

er
al

 p
op

ul
at

io
n,

 
Co

pe
nh

ag
en

, D
en

m
ar

k 

-S
em

i-q
ua

nt
ita

tiv
e 

FF
Q

 
-F

oo
d 

lis
t 

m
od

ifi
ca

tio
ns

 
to

 v
al

id
at

ed
 D

an
ish

 F
FQ

 
– 

Di
et

, C
an

ce
r 

&
 H

ea
lth

 
st

ud
y 

 
-C

om
pl

et
ed

 a
t h

om
e 

-9
 fr

eq
ue

nc
y 

op
tio

ns
 

-O
ve

r 1
 y

ea
r 

-F
FQ

 
-3

m
 in

te
rv

al
 –

 1
st

 4
-d

ay
 fo

od
 

re
co

rd
 (3

 w
ee

kd
ay

s +
 1

 w
ee

ke
nd

) 
-4

m
 in

te
rv

al
 –

 2
nd

 &
 3

rd
 4

-d
ay

 
fo

od
 re

co
rd

s  
(3

 w
ee

kd
ay

s +
 1

 
w

ee
ke

nd
) 

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.2

8 
(p

ro
te

in
) –

 0
.6

3 
(s

uc
ro

se
) 

-M
icr

on
ut

rie
nt

 C
C 

ra
ng

e:
 0

.2
1 

(v
it 

E)
 –

 0
.5

9 
(v

it 
C)

  
 -C

or
re

ct
 cl

as
sif

ica
tio

n 
: 3

2 
(v

it 
A)

 –
 6

4%
 (r

ib
of

la
vi

n)
 

-G
ro

ss
 m

isc
la

ss
ifi

ca
tio

n 
: 0

 (s
uc

ro
se

, v
it 

A 
&

 E
, c

al
ciu

m
) –

 8
%

 
(p

ro
te

in
, c

ar
bo

hy
dr

at
e,

 ch
ol

es
te

ro
l) 

Bi
ng

ha
m

 
et

 
al

. (
19

97
) 

-1
27

F,
 5

0-
65

y 
-N

or
fo

lk
  g

en
er

al
 

po
pu

la
tio

n,
 U

S 

-6
1 

ite
m

 
se

m
i- 

qu
an

tit
at

iv
e 

FF
Q

  
-F

oo
d 

lis
t 

m
od

ifi
ca

tio
ns

 
to

 
US

 
Nu

rs
es

’ 
He

al
th

 
St

ud
y 

FF
Q

 
-C

om
pl

et
ed

 a
t h

om
e 

-O
ve

r 1
 y

ea
r  

 
-1

st
 4

-d
ay

 fo
od

 re
co

rd
  

-3
m

 in
te

rv
al

 –
 2

nd
 4

-d
ay

 fo
od

 
re

co
rd

  
-3

m
 in

te
rv

al
 –

 F
FQ

 co
m

pl
et

ed
 &

 
3r

d 
4-

da
y 

fo
od

 re
co

rd
  

-3
m

 in
te

rv
al

 –
 4

th
 4

-d
ay

 fo
od

 
re

co
rd

  

-M
ac

ro
nu

tr
ie

nt
 C

C 
ra

ng
e:

 0
.4

3 
(p

ro
te

in
) –

 0
.5

5 
(fa

t) 
-M

icr
on

ut
rie

nt
 C

C 
ra

ng
e:

 0
.3

9 
(p

ot
as

siu
m

) –
 0

.5
5 

(v
it 

A)
 

(a
lco

ho
l -

 0
.9

0)
  

 - C
or

re
ct

 cl
as

sif
ica

tio
n:

 3
5 

(p
ot

as
siu

m
) -

 4
7%

 (v
it 

A 
&

 fi
br

e)
 

(a
lco

ho
l 7

5%
) 



   
   

 
 

34
  

Re
fe

re
nc

e 
Po

pu
la

tio
n 

Ch
ar

ac
te

ris
tic

s  
FF

Q
 D

es
ig

n 
St

ud
y 

De
sig

n 
Fi

nd
in

gs
 

(fo
od

 re
co

rd
s =

 v
ar

ie
d 

w
ee

k 
da

ys
) 

-G
ro

ss
 m

isc
la

ss
ifi

ca
tio

n:
 0

 (a
lc

oh
ol

) -
 6

%
 (p

ot
as

siu
m

 &
 β

-
ca

ro
te

ne
) 

No
te

. C
C 

= 
Co

rr
el

at
io

n 
Co

ef
fic

ie
nt

; P
UF

A 
= 

Po
ly

un
sa

tu
ra

te
d 

Fa
tt

y 
Ac

id
; M

UF
A 

= 
M

on
ou

ns
at

ur
at

ed
 F

at
ty

 A
cid

; S
FA

 =
 S

at
ur

at
ed

 F
at

ty
 A

ci
d;

 N
SP

 =
 N

on
-s

ta
rc

h 
Po

ly
sa

cc
ha

rid
e;

 v
it 

= 
vi

ta
m

in
; d

 =
 d

ay
s; 

m
 =

 m
on

th
; y

 =
 y

ea
r; 

EA
D 

= 
En

er
gy

 A
dj

us
te

d 
Da

ta
.



35 
 

2.9. FFQ’s Available for Use in Adult Female New Zealanders 

It is important that the FFQ reflects the current availability of food and consumption patterns 

of the population, as these do change over time (Cade et al, 2002; Willet, 1998). From the 

literature search, there were only five FFQs found that had been developed to assess multi-

nutrient intake in NZ adults (Bell et al., 1999; Bolch, 1994; Metcalf et al., 1997; Sam et al., 

2012