![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
Nutrition Consultant, Kaneohe, HI 96744 and Departments of
*
Occupational Therapy and
Food Science and Human Nutrition, Colorado State University, Fort Collins, CO 80523
3To whom correspondence should be addressed.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
KEY WORDS: • food security • hunger • Hawaii • validity • humans
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
,
Life Science and Research Office (LSRO)4
1990
]. Food security status has been measured traditionally on
an ordinal scale, beginning with food security through various levels
of food insecurity, including severe hunger among children
(Radimer et al. 1992
, Wehler et al. 1992
). The importance of food security measurement was
recognized in The Ten-year Comprehensive Plan for Nutrition
Monitoring and Related Research Programs [U.S. Department of Health
and Human Services (USDHHS) 1993]. With this recognition, the Food and
Consumer Service, USDA and the Center for Disease Control and
Prevention, National Center for Health Statistics, USDHHS and other
food security experts oversaw the development of an ordinal scale and a
categorical measure of household food insecurity called the Core Food
Security Module (CFSM; Carlson et al. 1999
). A summary
report of this landmark research effort (Hamilton et al. 1997a
), a technical report outlining development of the
national food security measures, (Hamilton et al. 1997b
), and the report of Price et al. (1997)
document this research. The CFSM, depicted in Table 1
, is used to assess the extent and
severity of food insecurity of households in the last 12 mo due to
inadequate money for food.
|
|
Development of the CFSM scale measure using the Rasch Model. |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
). The simple Rasch model,
developed by George Rasch, is a methodology for constructing reliable
and internally valid measures (Rasch 1966
, Wright and Stone 1979
). Applied to the CFSM, the assertions of the
simple Rasch model are as follows: 1) the more food secure a
respondent, the more likely the person is to respond negatively to
easier items, i.e., those indicating food security would answer
"almost never" to question (Q)2 "worried that food would run
out"; and 2) food insecurity items (Q2–6) are more likely
to be answered affirmatively than the hunger items (Q7–16), i.e., more
respondents will respond affirmatively to the least severe Q2, than the
more severe Q10 in which the respondent indicates hunger.
Rasch computer programs such as FACETS or BIGSTEPS model these
assertions mathematically (Linacre 1994
,
Wright and Linacre 1991
). Rasch programs transform raw
item scores into equal-interval scales (Wright and Masters 1982
). The basic Rasch model estimates the log-odds
probability of a given score (Fisher 1993
) as follows:
 |
where Pni is the probability of an affirmative response from a respondent n on item i, Bn is the food security measure of respondent n, and Di is the item hunger severity calibration of item i.
Because the logits are equal units of measurement, they are additive.
Both item hunger severity and respondent scale measures are calibrated
on the same linear scale. The item calibration values represent the
position of the item along the constructed food security measurement
scale. As depicted in Table 1
, an item such as Q16, with a high
positive item calibration value (4.82), indicates a greater degree of
insecurity and hunger, whereas an item with a low negative calibration
value, i.e., Q2: -4.99, indicates more food security or a lesser
degree of food insecurity (Hamilton et al. 1997b
).
Similarly, for respondent scale measures, which pertain to the degree
of food security the respondent experiences, a higher number of
affirmative responses indicate greater household food insecurity or
hunger and will result in a higher positive placement on the food
security scale. Few affirmative responses indicate less food insecurity
and therefore a negative placement on the food security scale. A more
complete description of the technical aspects of CSFM scale measure,
including Rasch measurement, can be found elsewhere (Hamilton et al. 1997b
).
An inspection of the ordering of the items by calibration values can be
used to examine the conceptual validity of the scales. Those items,
i.e., Q2, Q3, which should be "easy or less severe" are at the food
secure end of the scale and those that are "harder or more severe"
should be at the more food insecure (hungry) end of the scale. We
expect at least 95% of our sample to demonstrate valid patterns of
response across the items. The item calibration values can also be
examined for gaps in the scale, which can result in less sensitive or
less reliable measurements (Fisher 1993
). Standard
errors of the item calibrations and respondent food security measures
provide an estimate of reliability.
Mean square residuals (MnSq) are used to assess the goodness-of-fit of
each item compared with the assertions of the Rasch model. MnSq are
ratios of the observed vs. the expected scores. The expected values of
the MnSq are 1.0. In the development of the CFSM, MnSq values > 1.2 were judged indicative of a poorly fitting or erratic item,
especially when values of t, the standard residual, were
2. Using this criterion, the item was targeted for removal from
the scale. MnSq values < 0.8 indicate that the item was redundant
or lacked variability with respect to the information it shares with
another item (Hamilton et al. 1997b
), particularly when
values of t, the standardized residual, are 
-2, (Wright and Stone 1979
). Redundant or poorly fitting items were also
targeted for removal from the scale. Items that are erratic are removed
because their failure to demonstrate goodness-of-fit to the Rasch model
provides objective evidence (despite theoretical assumptions to the
contrary) that the item does not measure the same unidimensional
construct as do the other items included in the test. Items that lack
variability or are redundant are removed because they do not contribute
to the differentiation of persons into varying levels of ability
(person separation).
An additional advantage to Rasch measurement is that calibration values
are calculated independently of the respondents tested, independently
of the questions asked and independently of missing data. Thus,
responses from households with and without children, who answer a
different number of questions can be assessed using the same scale.
Furthermore, the fit among respondents can also be assessed.
Respondents who have a pattern of response that differs significantly
from expectations will misfit the modeled expectations. That is, they
are judged objectively to fail to meet the expectations of the Rasch
model for valid patterns of responses across the items (e.g., a more
food secure respondent failing food insecure items or a more food
insecure person responding affirmatively to hunger items). Usually, a
misfit rate 5% of the sample is deemed acceptable
(Wright and Masters 1982
). Finally, missing data do not
create a barrier to comparable data analysis (Wright and Stone 1979
).
|
The Core Food Security Module. |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
). Goodness-of-fit statistics of the respondent measures
were not reported. Analysis of fit across subgroups, i.e., households
with children under 18, elderly households with no children and
households with neither elderly nor children (single or multiple
adults) indicated the CFSM was robust across diverse household types
(Hamilton et al. 1997b
). |
CFSM with Asian and Pacific Islanders. |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
). Therefore, no more than 2% of
the population who responded to the Current Population Survey were
Asians and Pacific Islanders. Frongillo (1999)
, although
supporting the construction of the CFSM, cautioned that studies with
subgroups of the population and studies to validate the use of the
instrument for monitoring in other than national contexts were
required. Furthermore, to our knowledge, research used to develop
conceptual models of food insecurity did not include the diversity of
Asian and Pacific Islander participants in their samples
(Radimer 1990
, Radimer et al. 1992
,
Wehler et al. 1992
).
Asians and Pacific Islanders are a diverse ethnic category encompassing
Japanese, Chinese, Koreans, Vietnamese, Cambodians, Laotians,
Filipinos, Hawaiians, Samoans, Tongans, Chamorros, Hmong and others.
The percentage of Asian and Pacific Islanders throughout the United
States has risen from 3% of the U.S. population in 1990, to 3.7% in
1996, and is expected to increase to 5.1% by 2010 and to 8.7% by 2050
(U.S. Department of Commerce 1996
). However, Asian and Pacific
Islanders comprise 
50% of the population of Hawaii (Department of Business, Economic Development and Tourism 1997
). Documented
differences in cultural patterns, beliefs associated with food and
coping behaviors among the major ethnic groups that reside in Hawaii
(Palafox and Warren 1980
) suggested that perceptions of
food insecurity were likely to vary among Asians and Pacific Islanders.
|
Previous work. |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
, Derrickson and Anderson 2000
.
Qualitative findings indicated that the CFSM was likely to be valid and
reliable with our target audience (Derrickson and Anderson 2000
). However, Q4 "We couldn’t afford to eat balanced
meals" was troubling for many respondents who asked, "What does a
balanced meal mean?" When asked in return what they thought it meant,
respondents described predominantly a meal with meat, starch and
vegetables—no fruit, and no dairy products. However, some respondents
could not define a balanced meal. Pilot findings with a sample of
charitable food recipients also indicated that the CFSM was likely to
be reliable and valid with Asians and Pacific Islanders in Hawaii,
despite weakness with the "balanced meal" question
(Derrickson 1999
, Derrickson et al. in press
). |
Objectives. |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
with a sample from Hawaii.
We originally hypothesized that the CFSM would not be an adequate scale
measure to use with ethnically diverse samples in Hawaii. We
specifically addressed reliability as evaluated by the SEM
of the estimated food security status of the respondents as well as
test retest reliability. The evaluation of validity focused on the
following: 1) internal scale validity of the constructed
scale, 2) person response validity of the respondents,
3) comparison of the item difficulty calibration hierarchy
to theoretical models of food insecurity and 4) stability of
the CSFM item hierarchy between the national and the Hawaiian samples. |
SUBJECTS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
To validate the full range of food insecurity in a state in which 9.2%
of the population is thought to have experienced some degree of food
insecurity (Hamilton et al. 1997b
), the following three
samples were surveyed (n = 1664): 1)
a convenience sample of 144 food pantry recipients thought likely to be
hungry; 2) a retest sample that included 61of the
initial 77 food pantry respondents who completed the CFSM a second
time; and 3) a statewide sample of 1469 respondents
gathered through the Hawaii Health Survey (HHS).
All data were collected in Hawaii between June and November 1998 using the same instrument and similar data collection methods. Before data collection, all participants confirmed verbal consent required by a university Human Subjects Review Committee.
Names and phone numbers of food pantry recipients were gathered from
three nonprofit charitable food providers on O’ahu (Derrickson 1999
). Data collection began with a pilot study in the summer
of 1998 of 77 food pantry respondents who completed the survey once; 61
(80%) of those respondents also completed the survey again, an average
of 11 d later. Data were gathered by interviewers experienced in
calling households with limited resources (Derrickson et al. 1995
, SMS 1992
). Standard telephone survey
methods were used to enhance response rates and minimize interviewer
bias (Lavarakas 1988
, SMS 1998
). To the
extent possible, retest interviews were conducted without knowledge of
the first data collection responses.
An additional 67 food pantry participants and HHS participants were
gathered from September through November 1998. The HHS is a telephone
interview survey of 3500 households each year. It is modeled after
the National Health Interview Survey conducted by the National Center
for Health Statistics (SMS 1998
). Households were chosen
randomly from local phone books. Oversampling of households residing in
the counties of Maui and the Big Island of Hawaii was conducted to
further study households in these districts. Once a household was
chosen, the household was sent a letter from the Director of the
Department of Health encouraging survey participation. Data collection
of the remaining 67 food pantry respondents and data for all of the HHS
respondents was administered by phone interview using a
Computer-Assisted Telephone Interviewing system (SMS 1998
). A complete description of the data collection methods
used in the HHS are described elsewhere (SMS 1998
).
Survey instrument.
We used the guide created by Price et al. (1997)
to
direct data collection and analysis. Basic demographic information
(sex, household composition and ethnic disposition) was ascertained
before the food security questions. Thus, food security questions that
did not apply to households without children were not asked, and the
terminology "I" or "We" was used appropriately. The question
"With what ethnic group do you identify with most?" was
used to assess ethnicity. A total of 19 ethnic response categories were
collected, including one for no response and another for "mixed"
ethnicity.
The 18 food security questions were preceded by the four-part food
insufficiency question (Price et al. 1997
; Rose et al. 1995
). Exact wording of the questions and responses was
maintained and suggested "skip patterns" were employed to decrease
response burden (Price et al. 1997
). However, all
respondents were asked the food insufficiency question and at least Q2,
Q3 and Q4. Questions pertaining to use of various coping behaviors
(Hamilton et al. 1997b
), use of assistance programs,
income-related indices and dietary indices were completed after the
food security questions. Findings pertaining to additional data
collected have been reported elsewhere (Derrickson 1999
,
Derrickson et al. 2000a
and 2000b
). In the HHS, food
security questions were asked after other behavioral questions, but
before seeking more in-depth responses on income and other
demographics (SMS 1998
).
Data analysis.
For final analysis, food security responses were coded as 0 = negative response and 1 = affirmative response (Price et al. 1997
). However, instead of assuming negative responses for
questions that were not answered because the participant was
"screened out," we left responses to questions that were not asked
blank to avoid making incorrect assumptions about the data. The ability
to handle such missing data and avoid such error is an advantage of
Rasch measurement models. However, if a participant responded
negatively to a question with a temporal duration follow-up
question, i.e., "how often did this happen?" (Q 8–8a, Q 12–12a,
and Q 15–15a), a negative response was assumed for the follow-up
question to standardize the number of respondents to these pairs of
questions. These decisions were made after preliminary data analysis
was completed because this process most accurately reflects the data
and does not affect Rasch scale measures.
After preliminary assessment of the frequency responses, the following
indices or reclassifications were completed to assist in data analysis.
Ethnic classification were grouped into one of eight categories:
Hawaiian or part-Hawaiian, Caucasian, Filipino, Japanese, Other
Asian (Chinese and Mixed Asian), Samoans, Mixed or Unidentified, and a
combined category of African-Americans, Hispanics and Native
Americans. Using the algorithm of CFSM categorical measure
(Price et al. 1997
), the sum of affirmative responses
and household description (with or without children), each respondent
was classified into one of the following four household food security
categories: food secure, food insecure without hunger, food insecure
with moderate hunger and food insecure with severe hunger.
Analysis was completed with the FACETS Rasch computer program
(Linacre 1994
). The final sample used for Rasch analysis
comprised the 362 respondents who responded affirmatively to one or
more items. The 1300 who had no affirmative responses and two who had
no negative responses were automatically removed by FACETS from the
data set because they yielded no useful information for scale
validation purposes. Inclusion of the 61 food pantry survey responses
that comprised the retest sample was not viewed as a threat to validity
(Wright and Stone 1979
). Results without the 61 retest
respondents were similar to those presented herein (Derrickson 1999
).
Because concurrent presentation of findings and methods used to assess
internal scale validity, person response validity and reliability of
the measure is the most succinct approach, they are described
conjointly in the next section. Paired comparison Student’s
t test analysis and Pearson’s correlation analysis used
in stability assessment were completed using SPSS (Version 6.2, SPSS,
Chicago IL). The 
-value was set at 0.05.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Table 2
depicts the household
and ethnic description of the total sample. Of the 1664 people
surveyed, 999 (54.6%) indicated that they identified most with an
Asian or Pacific Islander ethnic group; 954 (57.3%) were from
households without children. The food pantry sample consisted of more
Hawaiians than the HHS sample (41.0 vs. 14.0%, respectively), more
Samoans (7.0 vs. 0.5%), more families (75.0 vs. 38.0%), fewer food
secure respondents (25.0 vs. 93.2%) and more female respondents (80.5
vs. 58.2%). The pantry and retest samples were quite similar except
that Samoans comprised a greater percentage of the retest sample (6.9
vs. 13.1%). Overall, 1411 (84.8%) were classified by the CFSM
categorical measure as food secure, 158 (9.5%) as food insecure
without hunger, 64 (3.8%) as food insecure with moderate hunger and 31
(1.2%) as food insecure with severe hunger.
|
To evaluate internal scale validity, we examined a number of variables.
First, we examined the goodness-of-fit of the items to the expectations
of the Rasch model (Table 3
).
Q8 and Q8a "adults cut the size or skip meals/often" had infit and
outfit MnSq values < 0.7 and outfit t-score values -2. Q4 "(un)able to eat balanced meals" had outfit MnSq values
>1.2 and t-values 2. We therefore concluded that these items
failed to meet our criteria for goodness-of-fit. The item separation
index of 9.29, calculated by dividing the adjusted standard deviation
of 2.29 by the real mean SEM of 0.24, indicated adequate
separation of items.
|
, we compared the national
scale of food security item calibration values (Table 1
;
Hamilton et al. 1997b
) with the Hawaii scale food
security item calibration values (Table 4
). The respondent measures of the
Hawaiian sample are depicted in the middle of Figure 1
. The value shown
by an asterisk in Figure 1
represents four participants, whereas a dot
(·) indicates one participant. As depicted in Table 4
, to determine
whether there were significant differences between the two sets of item
calibration values, we calculated the standardized difference (Z). A
Z-value 2 indicates a significant difference. As shown in Table 4
and illustrated in Figure 1
, t-values for Q2, Q3, Q4, Q8, Q8a,
Q11, Q15 and Q15a were significantly different. The range of the
national scale was greater than that of the Hawaiian scale. As a
result, items Q2, Q3, Q4, Q8 and Q8a were relatively less likely to
elicit a negative response (indicating food insecurity), and Q11, Q15
and Q15a were more likely to elicit a negative response among the
Hawaiian sample compared with the national sample. There were also
large gaps between the item calibration values within the data for
Hawaii. Most notably, the gap between Q2 and Q3 was 0.75 logit, and the
gap between Q4 and the next most severe item answered by all
households, Q9, was 1.80 logits.
|
|
To examine person response validity, we examined goodness-of-fit of the
respondents to the expectations of the Rasch model. Our criterion for
acceptable goodness-of-fit was similar to that used previously for item
fit. Seventeen people (4.7%) "misfit" or had MnSq values >1.2 and
z 2. An acceptable rate of misfit is 5%. As indicated in
Table 5
, although
there were no apparent differences in fit by site of the sample or by
household type, 5 of the 17 misfitting persons were Samoan.
|
To examine reliability, we examined the percentage of the total number
of item ratings that misfit, the test-retest correlation
coefficients between respondent scale measures at times 1 and 2 and the
respondent separation index. We found the percentage of item ratings
that misfit rate was 4.1% (186 of 4542 measurable responses). On the
basis of an expected 5% rate of misfit, we concluded that the items
were scored reliably in a manner expected by the Rasch model. Q4
"unable to eat balanced meals" was the only item to have
unacceptable reliability with a misfit rate of 6.7% (24 of 357
measurable responses). Also, the respondent separation index, an index
of the adjusted standard deviation of response measures to the real
SEM of response measures (1.76/1.17) of 1.51, indicated
that person response measures could be split reliably into only two
categories (Wright and Masters 1982
). [For further
discussion on Rasch approaches to the examination of reliability see
Fisher (1993)
, Wright and Stone (1979)
,
Wright and Masters (1982)
].
To assess the stability of the scale measure over time we compared the mean respondent scale measures of the 55 respondents who had scale measures at two times. Scale measures of -1.22 ± 2.04 logits at time one and -1.22 ± 2.01 logits at time two were not significantly different over time (t = -0.2, df = 54, P = 0.98). The Pearson product moment correlation coefficient of the respondent scale measures over time was r = 0.75 (P < 0.01).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
). Validity also
includes goodness-of-fit between an operational definition and the
response patterns of respondents to the items that comprise the
operational definition. The definition of food insecurity used by the
federal government is a condition that "exists whenever the
availability of nutritionally adequate and safe foods or the ability to
acquire acceptable foods in socially-acceptable ways is limited or
uncertain" (LSRO 1990
). This definition, the
operational definition of food security (Bickel et al. 1996
), and the responses of a sample of residents from Hawaii
who demonstrated evidence of food insecurity or hunger, provide the
basis for judging the validity of the CFSM scale measure in this study.
Based solely on the LSRO definition of food insecurity, it is
immediately apparent that the CFSM scale measure captures only the
"availability of nutritionally adequate food" aspect of this
definition of food insecurity, not the safety of foods or the social
acceptability of food acquisition methods. Overall internal scale construct validity and reliability.
Findings related to goodness-of-fit of the items suggest that overall,
the CFSM scale defines a single construct at least as well with
Hawaiian residents as it did in the national sample (Hamilton et al. 1997b
). Similar questionable fit statistics indicating
redundancy between Q8 "adults skip or cut the size of meals"
(Outfit MnSq = 76, Z = -4.6) and the follow-up question
Q8a. "How often" (Outfit MnSq = 0.77, Z = -2.6) were
noted in the original national fit (Hamilton et al. 1997b
). Similarly, high outfit statistics of Q2
"worried" (3.04: Z = 9.4), and Q4 (1.61: Z = 7.9) were
also noted in the national fit (Hamilton et al. 1997b
).
That is, the threats to the unidimensionality, observed through the
redundancy and relatively poor fit of a few items, were comparable
between the Hawaiian and national samples, These finding may suggest a
potential limitation of the CFSM scale measure that could have
practical applications for food security monitoring because the CFSM
categorical measure relies on reliable responses to all items for
appropriate food security status categorization (Hamilton et al. 1997b
, Derrickson et al. 2000b
).
The finding that only 4.7% of the participants misfit suggests that
the majority of participants who responded were evaluated in a valid
manner, consistent with the expectations of the Rasch model. Table 4
also indicates that the rate of misfit among households of different
composition, site of sample and food security status were consistent
with the proportion of respondents in these groups. However, with a
total of 23 Samoan respondents, the existence of five (21.7%) misfits
among this ethnic group, which is greater than the 5% that would be
expected to misfit by chance, suggests an inadequate fit of the CFSM to
the Samoans in this sample. However, sample size limits our confidence
in this conclusion as does the relatively higher number of Samoans who
were classified as experiencing either hunger among adults and/or
children (47% Samoans vs. 26% overall, n = 362).
Thus, it is unclear whether the high rate of misfit is because of an
ethnic difference in reporting of the Samoans, or because the Samoans
were more likely to demonstrate different patterns of food insecurity.
Findings also indicated an acceptable level of stability of the CFSM
scale measure over a mean of 11 d. McGuiness (1996)
also found acceptable stability of the CFSM with a national sample.
Thus, overall findings indicate that, except for the Samoans, the CFSM
scale measure is as reliable and valid with Asians and Pacific
Islanders in Hawaii as it was in the national sample (Hamilton et al. 1997b
).
Q4 "(un)able to eat balanced meals."
When we examined the items associated with a higher percentage of
misfitting individuals and/or unacceptable goodness-of-fit statistics,
we found Q4 "(un)able to eat balanced meals" to be ambiguous, with
responses that are likely to cause random errors and lower response
rates. Our previous work indicated there were different interpretations
of the meaning of this question (Derrickson and Anderson, 2000
), and a relatively low correlation of responses to Q4 over
time (r = 0. 3, P = 0.04, n
= 59) (Derrickson 1999
). Thus, findings confirm
that inconsistent understanding of the term "balanced meals" is
likely affecting the reliability and the validity of responses to Q4.
On the basis of additional qualitative work in Hawaii, we have
suggested rephrasing this question to "Unable to afford to eat a meal
containing starch like bread or rice (or appropriate starch), a
protein-rich food like meat, milk, fish or beans, and a fruit or a
vegetable" (Derrickson et al. 2000
).
Respondent response validity and reliability.
If the purpose of the CFSM "is to accurately identify the extent and
severity of food insecurity of the respondents" (Carlson et al. 1999
), findings outlined in this study indicate that the
CFSM scale measure may not be adequate in differentiating food security
from relatively mild food insecurity for the following reasons:
, in which a relatively large gap in
the item calibration values occurs between Q2 "worried" (-4.18)
and Q3 "food did not last" (-3.43).
When a test is developed, the developer should attempt to
conceptualize a construct and then develop items that are related to
that construct. An important aspect of this, which is recognized in
Rasch measurement but often overlooked when using traditional models,
is that the range of the difficulty of the items should parallel the
ability of the people to be tested (Wright and Masters 1982
, Wright and Stone 1979
). In common
measurement terms, there is a need for easy items for less able people,
and difficult items for more able people. Although there is a need for
more items on the scale in which decisions are made, there is also a
need for sufficient items along the remainder of the scale so as to
adequately spread the people out into differing levels of ability or in
this case, food security status. Although conceptual-based item
development is important initially, data-based validation of that
theoretical model is also required. Items thought to be good
discriminators conceptually can be found not to be effective in
practice, and in the case of the CSFM, there are sufficient gaps to
question whether discrimination is occurring.
The need for additional items is also exemplified in
criterion-mediated validity assessments between the CFSM scale
measure and related variables. Rasch methods are able to estimate scale
measures for respondents who answer all questions affirmatively or who
do not answer any questions asked of them affirmatively. However, these
estimates are associated with very large standard errors of
measurement, making their validity and reliability of questionable use
in practice. In a study such as this one, persons who obtain maximum or
minimum total scores also do not contribute to the validation of the
scale. For these reasons, in this study, out of the 1664 possible
respondents, responses from only 362 (22%) were available for Rasch
analysis. Thus, criterion-mediated validity assessment was limited
only to these 362 rather than all 1664 respondents. Given that
criterion-mediated validity cannot be confirmed confidently or
understood without accurate comparisons between the most and least food
secure (Frongillo et al. 1997
, Kendall et al. 1996
, Tarasuk and Beaton 1999
), use of the CFSM
scale measures in concurrent validity assessments without additional
items confirming food security status appears limited
(Derrickson et al. 2000a
). If the scale measure is to be
considered the "standard reference measure" of food security status
using LSRO definitions of food security status, then additional items
indicative of food security and mild food insecurity may have to be
included.
Practical applications to food security monitoring and research.
The practical applications of previously identified limitations of the
CFSM are manifested in the CFSM categorical measures and shorter food
security scales (Blumberg et al. 1999
). In the
CFSM categorical measure, the measure currently used to assess national
food security status (Nord et al. 1999
), three
affirmative responses are required for classification as food insecure
(Hamilton et al. 1997b
, Price et al. 1997
). Because Q3, Q4 and Q5 all cluster quite tightly within
0.75 of a logit of each other, it appears as though the conceptual
basis of the measure supports affirmative responses to these measures
as clearly a more severe phenomenon than Q2 "worried food would not
last." Yet, is not clear why a single response indicating "worried
that food would run out," which appears consistent with the
definition of food insecurity (LSRO 1990
), or two
affirmative responses are categorized as food secure rather than food
insecure. We believe these issues warrant further investigation and
have begun this exploration in a subsequent research project
(Derrickson et al. 2000b
).
Researchers at the National Center for Health Statistics investigated
using a set of six items that include questions Q3, Q4, Q8, Q8a, Q9 and
Q10 (Blumberg et al. 1999
) as an alternative to the
18-question CFSM categorical measure for food security monitoring. Item
and respondent validity and reliability of the CSFM scale measure are
paramount if it is used as the basis of smaller food security
monitoring instruments. Given that the item calibration of Q8 in the
Hawaii sample was > 2.0 logits different than from the original
national sample (-0.78 Hawaii and -1.72 national), that Q8 and Q8a
were redundant in both samples, and the previously mentioned issues
with Q4, dependence on these three items (without rewording Q4) in a
subscale of only six items appears problematic. Although we support the
use of a shorter food security measure for food security monitoring
purposes (Derrickson et al. 2000b
), on the basis of the
findings herein, use of the proposed six subscales for food security
monitoring or research appears premature at this time.
Specific recommendations for use of the CFSM.
We used the published guidelines to direct our data collection
(Price et al. 1997
). These guidelines suggest
utilization of "skip patterns," such as not having households
continue the survey if they responded negatively to all previously
answered questions. On the basis of our experience, we would recommend
that those interested in basic research with the CFSM not use skip
patterns because the pattern of responses is variable, and the use of
skip patterns limits measurement of variable responses and requires
different assumptions. If the skip patterns are employed to minimize
data assumptions, we recommend that data input for missed questions
should be blank rather than zero for a negative response. However, for
applied uses, we believe that the decreased response burden, which
lowers cost and reduces interviewer and respondent fatigue, justifies
the use of skip patterns. Further research is required to compare
methods of CSFM administration and data analysis.
Limitations.
Reference to the robustness of the instrument among Asian and Pacific Islanders is limited to Hawaii. Given the myriad of Asian and Pacific Islander groups and differences in acculturation within groups, additional studies are required before any conclusions on the robustness of the CFSM with different ethnic groups can be made. The small number of Samoans sampled in this study limits the certainty of any conclusions regarding the fit of the CSFM with this ethnic group. Additional research work, with Samoans of varying degrees of "Westernization," i.e., from Western Samoan, American Samoa and Samoan-Americans, is warranted.
Implications.
This is the first study to validate independently the internal scale
validity and reliability of the CFSM scale measure with an ethnically
diverse state sample. Findings suggest that the CSFM scale measure
demonstrated internal scale construct validity, person-response
validity, stability and an item hierarchy consistent with conceptual
expectations (Bickel et al. 1996
, Radimer et al. 1992
) and previous work (Hamilton et al. 1997b
).
Preliminary findings and implications that were shared at the Second
Food Security Measurement and Research Conference (Derrickson et al. 2000c
) have now been identified in part as
priority areas for food security research (Economic Research Service 1999
). With the exception of Samoans, our findings
suggest a promising "ethnic" robustness of the CFSM among the
ethnic groups studied in Hawaii, at least to the extent that the CFSM
is valid nationally. These findings support the potential application
of the CFSM to measure the extent and severity of food insecurity among
various ethnic groups throughout the United States and in samples that
are more diversified than national samples. However, identified
weakness in the 18-item scale, gaps in the scale and the poor fit of
Q4, Q8 and Q8a are important when considering the validity and utility
of the scale measures for research and the categorical measure for food
security monitoring. Revision of the wording of Q4 (Derrickson et al. 2000
), and the addition of new food security and insecurity items were
suggested for consideration. Given the significance of application of
the CFSM, we caution that before implementing any changes in the CFSM
scale of indicators, the CSFM categorical measure or the use of food
security subscales, additional validation studies with diverse food
insecure populations be completed.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
2 Funded in part by a grant from the Institute for
Research on Poverty, University of Wisconsin, Madison, WI.
4 Abbreviations used: CSFM, Core Food Security
Module; HHS, Hawaii Health Survey; LSRO, Life Science and Research
Office; Mn Sq, mean-square residuals; Q, question; USDHHS, United
States Department of Health and Human Services.
Manuscript received October 8, 1999. Initial review completed December 17, 1999. Revision accepted May 31, 2000.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONDevelopment of the CFSM...The Core Food Security...CFSM with Asian and...Previous work.Objectives.SUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
1. Bickel G. Toward a research agenda: next steps 1999 Second Food Security Measurement and Research Conference Alexandria, VA.
2. Bickel G., Andrews A., Klein B. Measuring food security in the U.S.: a supplement to the CPS. Hall D. Stavrianso M. eds. Nutrition and Food Security in the Food Stamp Program 1996:91-111 USDA Food and Consumer Service Alexandria, VA.
3.
Blumberg S. J., Bialostosky K., Hamilton W. L., Briefel R. R. The effectiveness of routine measure of financially based household food insecurity. Am. J. Public Health 1999;89:1231-1243
4. Carlson S. J., Andrews M. S., Bickel G. W. Measuring food insecurity and hunger in the United States: development of a national benchmark measure and prevalence estimates. J. Nutr. 1999;129:510S-516S
5. Center for Nutrition Policy and Promotion (USDA) Nutrition action themes for the United States. A report in response to the international conference on nutrition. CNPP–2, Occasional Paper 1996 Washington, DC.
6. Department of Business, Economic Development and Tourism, State of Hawaii State of Hawaii Databook 1997:1995 Honolulu HI.
7. Derrickson J. P. Independent Validation of the Core Food Security Module with Asians and Pacific Islanders. Doctoral thesis 1999 Colorado State University Fort Collins, CO.
8. Derrickson J. P., Anderson J. A. Face validity of the core food security module with Asians and Pacific Islanders. J. Nutr. Ed. 2000;32:21-30
9. Derrickson J. P., Anderson J. A., Fisher A. Assessment of food insecurity among Asians and Pacific Islanders 2000c Second Food Security Measurement and Research Conference Alexandria, VA in press
10. Derrickson J. P., Anderson J. A., Fisher A. G. Concurrent-validity of a face valid food security measure 2000a University of Wisconsin Institute for Research on Poverty, Discussion Paper no. 1206–00, http//www.ssc.wisc.edu/irp/dplist.htm.
11. Derrickson, J. P., Anderson, J. A. & Fisher, A. G. (2000b) An assessment of various household food security measures in Hawaii has implications for national food security research and monitoring. J. Nutr. (submitted for publication #99–6558).
12. Derrickson J., Maeda I., Sonomura S., Braun K. Nutrition knowledge and behavioral assessment of participants of aid for families with dependent children: telephone vs. mail data collection methods. J. Am. Diet. Assoc. 1995;95:1154-1155[Medline]
13. Derrickson, J. P., Saka M. & Anderson, J. A. Interpretations of the "Balanced Meal" Household Food Security Indicator. J. Nutr. Ed. (in press).
14. Economic Research Service (USDA) (1999) Food security: measurement and research priorities identified. In: Second Food Security Research and Measurement Conference. http://www.econ.ag.gov/briefing/foodasst/fsresearch.htm.
15. Fisher A. G. The assessment of IADL motor skills: an application of many faceted Rasch analysis. Am. J. Occup. Ther. 1993;47:319-329[Medline]
16. Frongillo E. A. Validation of measures of food insecurity and hunger. J. Nutr. 1999;129:506S-509S
17.
Frongillo E. A., Rauschenbach B. S., Olson C. M., Kendall A., Colmenares A. G. Questionnaire-based measures are valid for the identification of rural households with hunger and food insecurity. J. Nutr. 1997;127:699-705
18. Hamilton, W. L., Cook, J. T., Thompson, W. W., Buron, L. F., Frongillo, E. A., Jr., Olson, C. M. & Wehler, C. A. (1997a) Household Food Security in the United States in 1995: Summary Report of the Food Security Measurement Project. Report prepared for the USDA, Food Consumer Service, Alexandria, VA.
19. Hamilton, W. L., Cook, J. T., Thompson, W. W., Buron, L. F., Frongillo, E. A., Jr., Olson, C. M. & Wehler, C. A. (1997b) Household Food Security in the United States in 1995: Technical Report of the Food Security Measurement Project. Report prepared for the USDA, Food Consumer Service, Alexandria, VA.
20. Kendall A., Olson C. M., Frongillo E. A. Relationship of hunger and food insecurity to food available and consumption. J. Am. Diet. Assoc. 1996;96:1019-1024[Medline]
21. Lavarkas P. J. Telephone Survey Methods: Sampling, Selection and Supervision 1988 Sage Publications Newbury Park, CA.
22. Life Sciences Research Office (Anderson S. A. Core indicators of nutritional state for difficult-to-sample populations. J. Nutr. 1990;120:1557S-1600S
23. Linacre J. FACETS 1994 MESA Press Chicago, IL.
24. McGuiness, R. (1996) Response variance in the 1995 food security supplement: reinterview report. Quality Assurance and Evaluation Branch, Demographic Statistical Methods Division, U.S. Bureau of the Census, Washington, DC.
25. Nord M., Jemison K., Bickel G. Prevalence of food insecurity and hunger by state 1999:1996-1998 Food and Rural Economics Division Economic Research Service, U.S. Department of Agriculture, Food Assistance and Nutrition Research Report No. 2, Washington, DC.
26. Palafox N. Warren A. eds. Cross-Cultural Caring 1980 A Handbook for Health Care Professionals in Hawaii. Transcultural Health Care Forum Honolulu, HI.
27. Price C., Hamilton W. L., Cook J. T. Household food insecurity in the United States: guide to implementing the Core Food Security Module 1997 Food and Consumer Service, United States Department of Agriculture Alexandria, VA.
28. Radimer K. L. Understanding Hunger and Developing Indicators to Assess It 1990 Doctoral thesis Cornell University, Ithaca, NY.
29. Radimer K. L., Olson C. M., Greene J. C., Campbell C. C., Habicht J. P. Understanding hunger and developing indicators to assess it in women and children. J. Nutr. Ed. 1992;24:36S-45S
30. Rasch G. An item analysis that takes individual differences into account. Br. J. Math. Stat. Psych. 1966;4:321-333
31. Rose D., Basiotis P. P., Klein B. W. Improving federal efforts to assess hunger and food insecurity. Food Rev. 1995;:18-23
32. Singleton R. A., Straits B. C., Striats M. M. Approaches to Social Research 2nd ed. 1993 Oxford University Press New York, NY.
33. SMS Research and Marketing Service, Inc Homelessness and hunger in Hawaii 1992 Presented to homeless Aloha, June 15, 1992 Honolulu, HI.
34. SMS Research and Marketing Service, Inc Hawaii Department of Health, Office of Health Status Monitoring 1998 Procedure Manual, Hawaii Health Survey-1997 Honolulu, HI.
35.
Tarasuk V., S & Beaton G. H. Women’s dietary intakes in the context of household food insecurity. J. Nutr. 1999;129:672-679
36. U.S. Department of Commerce, Economics and Statistics Administration, Bureau of the Census (1996) Resident population of the United States: middle series projections, 1996–2000, by sex, race and Hispanic origin, with median age. http://www.census.gov/population/projections/nation/nsrh/nprh 9600.txt.
37. United States Department of Heath and Human Services, Department of Agriculture Ten-year Comprehensive Plan for the National Nutrition Monitoring and Related Research Program 1993 Federal Register, Friday, June 11, 1993, 58 32 753–806.
38. Wehler C. A., Scott R. I., Anderson J. J. The community childhood identification project: a model of domestic hunger—demonstration project in Seattle, Washington. J. Nutr. Ed. 1992;24:29S-35S
39. Wright B. D., Linacre J. M. BIGSTEPS 1991 Rasch Analysis Computer Program. MESA Press Chicago, IL.
40. Wright B. D., Masters G. N. Rating Scale Analysis 1982 MESA Press Chicago, IL.
41. Wright B. D., Stone M. H. Best Test Design 1979 Rasch Measurement. MESA Press Chicago, IL.
This article has been cited by other articles:
|
|
 |
|
  M. Hackett, H. Melgar-Quinonez, R. Perez-Escamilla, and A. M. Segall-Correa Gender of respondent does not affect the psychometric properties of the Brazilian Household Food Security Scale Int. J. Epidemiol., August 1, 2008; 37(4): 766 - 774. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Isanaka, M. Mora-Plazas, S. Lopez-Arana, A. Baylin, and E. Villamor Food Insecurity Is Highly Prevalent and Predicts Underweight but Not Overweight in Adults and School Children from Bogota, Colombia J. Nutr., December 1, 2007; 137(12): 2747 - 2755. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. R. Melgar-Quinonez, A. C. Zubieta, B. MkNelly, A. Nteziyaremye, M. F. D. Gerardo, and C. Dunford Household Food Insecurity and Food Expenditure in Bolivia, Burkina Faso, and the Philippines J. Nutr., May 1, 2006; 136(5): 1431S - 1437S. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Perez-Escamilla, A. M. Segall-Correa, L. Kurdian Maranha, M. d. F. A. Sampaio, L. Marin-Leon, and G. Panigassi An Adapted Version of the U.S. Department of Agriculture Food Insecurity Module Is a Valid Tool for Assessing Household Food Insecurity in Campinas, Brazil J. Nutr., August 1, 2004; 134(8): 1923 - 1928. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. D. Opsomer, H. H. Jensen, and S. Pan An Evaluation of the U.S. Department of Agriculture Food Security Measure with Generalized Linear Mixed Models J. Nutr., February 1, 2003; 133(2): 421 - 427. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. P. Derrickson, A. G. Fisher, J. E. L. Anderson, and A. C. Brown An Assessment of Various Household Food Security Measures in Hawaii Has Implications for National Food Security Research and Monitoring J. Nutr., March 1, 2001; 131(3): 749 - 757. [Abstract] [Full Text]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
