Past CDPS reports have shown that lower acculturation is associated with higher levels of intake of fruits and vegetables among Latinos.This raises the question if seasonal variation affects high and low acculturated Latinos differently. The modified 24-hour recall method used in the CDPS requires considerable effort and resources to implement correctly. A significant challenge exists in training “generic” commercial interviewers who generally do not have a nutrition background. Interviewers play a critical role in assisting respondents in assessing the number of servings correctly for different reported items . They also must aid respondents in deconstructing mixed dishes to determine the number of servings of the different components. Not all interviewers are able to master this equally and over the course of weeks of data collection there is degradation of knowledge due to a combination of rare occurrences of some food items and forgetfulness. As a result, significant effort is required for quality control monitoring of interviewers and periodic refresher training to make the CDPS method work. In implementing the CDPS method, the length of time used in one interview is dependent on 1) what the respondent understands, 2) how extensive or varied the respondent’s fruit and vegetable consumption is, and 3) the skill of the interviewer. A good interviewer could normally accomplish this task in five to seven minutes. Time, however, is an important factor in whether or not general health behavior surveys can afford to include multiple complexquestions on dietary intake,plastic planter pot especially in determining the number of servings of fruit and vegetables consumed.
A shorter form of the CDPS method that is less involved and less complex with regard to interviewer training would be a convenient substitute if it worked as well in producing population estimates. Such an approach, derived from the CDPS method and asking only three questions, is compared to the CDPS method as part of this study. The possible substitute method is called the “Short Form 3” or “SF3,” since it uses three short and direct questions . The study population is California adults, ages 18 years and older. Among these persons, those who self-identify as White, Latino, or African American and those who report that their total annual household income from all sources is $25,000 or less are oversampled. A dual frame design is used to locate these persons. The method of data collection is the same used in the CDPS, a telephone survey using computer-assisted telephone interviewing techniques. The main frame consists of all residential telephone numbers across the state. The second or supplemental frame consists of residential telephone numbers located in geographic areas with concentrations of Latino, African-American, and low-income households. Interviews completed in this supplemental frame are classified as “targeted,” since they are designed to maximize the chance of reaching the groups of interest. Sample sizes were designed to deliver an estimated sampling precision with a statistical power of 0.80 to discriminate with 95% confidence between groups if four-month seasons were defined and used . The combined four-month period was originally chosen a priori for design purposes. A single month is the minimum time frame used, since it was felt that the data might reveal seasons different than the a priori seasons described in the original proposal. The “seasons” were originally defined as summer , winter , and spring . To provide maximum flexibility in examining seasonal variation, independent samples were selected for each month of the year.
The samples were further stratified across two years for two reasons: 1) to smooth out any inter-year variation and 2) to expense data collection within the annual funding cap. Since analysis would be at the month level, the calculated sample size for any given month was divided between the month in Year 1 and the same month in Year 2. This is a rectangular sample design in that the number of interviews would be completed for each of the three race/ethnic groups per month for both the general population and for the low-income population. The original calculation based on the hypothetical four-month season was for 660 completed interviews per race/ethnic group per month. This is 165 interviews per month, each month divided over two years so that 83 interviews are collected for each of the four months in Year 01 and the same number per month in Year 02. This calculated number allows for discrimination down to 0.55 servings using a standard deviation from the 1999 CDPS of 3.5 servings for all groups for the hypothetical four-month season. Examining groups by combining all 12 months would have still greater discrimination . Due to the higher cost per case in obtaining low-income race/ethnic-specific interviews, the goal for all three low-income groups was set to be 400 interviews per hypothetical four-month season. At the same confidence level and power , 400 cases discriminates differences greater than 0.69 servings. For 12 months of data combined, an n of 1,200 cases will discriminate differences greater than 0.4 servings. This is a monthly sample size of 51 cases per month per year for each of the three low-income race/ethnic groups. Most CDPS data collected in the past cover the above so-called summer months of July through October. The general population survey has never been collected in the December through May period with the exception of the last two days in May during the very first CDPS in 1989. Some African-American over-sample cases have been collected as late as November .
The sample provides an identical snapshot of the California population and the sub-groups of interest inside of each month. It is important to note that an individual case may be used to satisfy different sample size objectives. For example, a low-income African American case located in the general population random-digit dial survey is used in a) the general population estimate, b) the low-income population estimate, and c) the African-American estimate. Individuals located through the targeted over-samples are only used for their group member estimates. The data collection instrument used in this study is the fruit and vegetable intake module of the CDPS. Also included are the five language-based acculturation questions asked of all Latino respondents . Descriptive, self-reported demographic data are collected to define gender, education, race/ethnicity, and income. The SF3 questions are implemented in half of the study population by random selection. Using the short form in only half of the sample allows for measuring and adjusting for any potential testing effect of placing these questions ahead of the CDPS module. Placing it after the module, however, would be counterproductive since the CDPS module walks each respondent through each meal on the previous day and details specific servings consumed. This would greatly influence the SF3 response toward a higher level of agreement with the module and result in overstating the agreement of the SF3 estimates. Data collection was conducted using CATI methods. Forty interviewers were trained, ten of whom were bilingual Spanish-English. Each month’s sample was administered so that 80% of the cases were collected during the first three weeks of the month and the balance of the cases, including hard-to-reach cases and any remaining refusal conversions, occurred in the final week of the month. The objective was to have the interviews spread as evenly as possible over the entire month and all cases in the sample for a given month completed within that month. If any month fell short of the target number of cases in Year 1, the difference was made up in the same month in Year 2. To accomplish this, sample management in Year 2 was very exacting. Each month’s cases were the result of an independently drawn, random sample of California households. Respondents were then randomly selected from among all eligible respondents in the household. Thus,30 litre plant pots this is a two-stage random sample design. Only the selected respondent would or could be interviewed. Interviews were conducted in either English or Spanish at the preference of the respondent. At least one subsequent refusalconversion attempt was made in households that refused to participate. At least nine contact attempts were made on each selected telephone number. Interviews were completed between November 1, 2000 and October 31, 2002. The average interview took 9.4 minutes to complete. This is just under the 10 minutes originally planned and budgeted. Of the 8,614 completed interviews, 1,249 were completed in Spanish. The overall response rate* was 26.5% for the general population survey, the refusal rate* was 5.6%.
These rates were computed by the data collection vendor based on their available disposition coding scheme. Inadequate tracking of disposition codes by the vendor for the targeted and low-income samples made response rate calculations unreliable for these groups. This makes all disposition codes suspect and may account for the lower than expected general population response rate. The data file of final cases was cleaned by the vendor and the fruit and vegetable codes added to the recorded fruit and vegetables in each data record. The data collection vendor’s final report is included in Appendix III. As done in the CDPS, interviewers entered the actual fruit, vegetables, salad ingredients, and the fruit and/or vegetable mixed dishes reported by respondents. These standardized entries were post processed by the data collection vendor using programming that read and converted these alpha entries to the numeric codes used by the CDPS. These codes are based on, although not identical to, USDA food codes. These codes had been updated during the course of this study using the 2001 CDPS data. This work was done by one of this study’s research assistants who was also a registered dietitian and was reviewed by the principal investigator in collaboration with Public Health Institute staff who works with the CDPS for the California Department of Health Services. The number of servings of fruit and vegetables was recorded by interviewers as whole numbers. Respondents reported a serving size as what is “usual” for them. All reports of a half a serving or greater were rounded up to the next whole number. For all amounts greater than one serving where the respondent reported more than the whole number, but less than an additional half serving , the number was rounded downward. The exception is when the amount reported was less than one-half serving. In this instance, the interviewer entered a zero. This is particularly true for such items as lettuce and tomato on a sandwich or on a taco . This is consistent with the CDPS. This study differs from the CDPS in one respect. The analysis of the CDPS data recodes the zero entries as quarter servings, while this study did not. It is the opinion of both the authors that the relative relationship among the groups studied and among months sampled remains unchanged when not recoding the zero entries, thus the data are still valid for purposes of this study’s objectives. An examination of the number of reported servings revealed, as expected, cases with unusually high numbers of servings. After cases with likely recording errors are dropped , it is generally accepted to top code outlier cases. Consuming a high number of servings of fruit and vegetables may not be unusual, particularly for vegetarians. However, to minimize the impact of these outlier cases on the computed mean values and variance calculations, it is typical to top code these cases to a determined value. Initially, weexplored computing outlier cutoff values using the same method employed by diet researchers at the National Cancer Institute in their work with National Health Interview Survey data.The method involves identifying the first and third quartile in the study’s data distribution. This is done independently for fruit and for vegetables after transforming the variable by using the square root of the number of servings. The maximum value for the total number of servings of fruit and vegetables combined is computed to be 20.43 servings, rounded down to 20 servings. However, the CDPS has, with experience of over a dozen years, top coded the number of servings of fruit and the number of servings of vegetables each at 10.0 servings. Following that convention and thus staying consistent with CDPS, all outlier cases for servings of fruit and for vegetables in this study were top coded at 10 servings. No case can thus exceed 20 servings of fruit and vegetables combined, and this is, coincidentally, the same number of servings computed earlier following the NCI method.