Jill Boylston Herndon, Ph.D.
Institute for Child Health Policy
College of Medicine, University of Florida
I'd like to begin by thanking Melissa Riddle and Dave Clark at NIDCR for organizing this webinar series. A significant component of our work at the Institute for Child Health Policy is evaluating state Medicaid and CHIP programs, which frequently involves conducting surveys and analyzing survey data. The focus of my talk today will be on some key considerations when working with multiple survey data sources.
There are three main areas that we will address. The first involves considerations when comparing data on a similar domain from different surveys. Next, we’ll look at the limitations and opportunities for using national survey data to conduct state and local analyses. Finally, we will discuss some of the opportunities for linking state survey data to other data sources.
The motivation for this webinar comes from the various ways that people use survey data for conducting research and policy analysis. So, a lot of times we may be interested in comparing data on a given health domain from different surveys. For example, if you're conducting a state or local survey, you may want to compare your results with those from national surveys, or you may be interested in using data from different surveys in order to provide contextual information. There also is often interest in using national survey data to conduct state or local analyses or to inform state and local policymaking. You also may want to link survey data to other types of data, such as administrative data, in order to create richer analytic data sets.
However, these various data sources may not be directly comparable or easily connected, which has important implications for both the ability to conduct the desired analyses and the interpretation of results.
So, the purposes of this webinar are to provide an overview of the key considerations in comparing and linking survey data and to offer strategies and resources for working with different data sources.
For a variety of reasons there is often interest in using or comparing data from different surveys and there are certain health domains that are commonly included in national and state surveys.
For example, health insurance coverage is measured by several national surveys as well as many state surveys. There also are multiple national surveys that allow one to estimate the percentage of people who have received dental care. However, the estimates derived from the different surveys on each of these domains are different and sometimes the magnitude of the differences can be substantial. We’ll review four key considerations that survey users and researchers should take into account when analyzing and interpreting survey data, and then come back to these examples to better understand why estimates may vary across surveys.
The first consideration is the primary purpose of the survey. This may seem pretty fundamental but it's easy to get focused on the particular domains and data elements that you are interested in and lose sight of the larger context in which the data were collected. That larger context has significant implications for a number of factors that can influence how the domains of interest are measured. These factors include the target population for the survey. For example, is it working aged adults or does it also include children and individuals 65 years and older?
The primary purpose of the survey also affects how the respondent’s attention is focused. Attention may be more or less focused on the topics you're investigating. Moreover, it affects how in depth the domains of interest are covered.
In addition, the primary purpose of the survey affects the context in which questions are asked and their placement in the survey. Are the questions of interest placed early on, at the end of the survey, or somewhere in between?
The second consideration is the survey’s target population, which is the overall population to which the results are to be generalized. Key populations of interest to your study may be underrepresented or excluded because they were not the main target population for the survey. Examples include children, individuals with chronic conditions, and individuals within certain racial or ethnic groups.
It can also be helpful to understand the specific inclusion and exclusion criteria which may vary within the same overall target population across different surveys. For example, inclusion/exclusion criteria may vary with respect to Hispanic subpopulations; that is, the specific Hispanic and Latino groups that were included in the sample population. So, the category of “Hispanic” may not be comparable across surveys.
In addition to understanding what subpopulations were included in the data collection stage, it is also important to know what demographic and subpopulation information is included in the data files and how the variable categories are defined. For example, if there is a race variable, what are the specific categories within that variable and how do those categories compare across surveys? For continuous variables that get collapsed into brackets, for example, age brackets or income brackets, what are those brackets and how do they compare across surveys? These considerations affect both the comparability of data across the surveys as well as the ability to generalize to the subpopulations of interest.
The third consideration when using different surveys is how the survey methodology compares. The survey population and sampling frame delve into exactly how the target population will be represented in the survey. So, here we are thinking about the specific units that will be sampled. Is it a household, an individual, an organization, and so forth. The sampling frame is the list of sampling units from which the sample is selected and the sample design indicates how the sampling unit will be selected from the sampling frame, such as probability sampling or nonprobability sampling.
When examining the design of national surveys is important to be aware of the populations and geographic areas for which the survey was designed to produce reliable estimates. The survey or sample population represents an operational definition of the target population. Frequently, this is the accessible population within the target population. The accessible population is related to the survey mode. For example, a mail survey requires valid address information. A phone survey requires a working phone number. So, individuals within the target population without valid contact information would be excluded from the survey population.
Understanding the survey methodology is important because it influences response rates and the representativeness of the sample. In addition to influencing response rates, the survey mode may also affect the responses themselves. For example, a study that compared different survey data collection methods for obtaining substance use information found that respondents to a mailed survey were more likely to report substance use than those responding to a telephone interview.
When comparing data on the same health domain across different surveys, or even for the same survey over time, it is important to compare the question phrasing and placement. Information on the specific health behavior may be elicited in different ways across surveys. But even subtle differences in question wording may influence how the survey respondent answers the question. Question wording may influence what is referred to as the social desirability effect, where respondents may consciously or unconsciously answer the question based on their perception of social norms or expectations rather than based on their actual behavior.
Question wording includes, of course, the response options. Two surveys may have an identically worded question but the response options to that question may vary. Differences in question wording or the response options can affect the comparability across surveys.
Questions about health behaviors often include a time reference but the reference period may not be the same across surveys. In general, the longer the reference period, the more likely it is that there will be reporting inaccuracies. In addition, the less clearly defined is the time reference period, the more it is subject to varying interpretations by different respondents which also contributes to less accurate reporting.
As noted earlier, where the questions of interest are placed in the survey and the overall context in which they are asked can influence how the respondent’s attention is focused, and therefore may influence reporting accuracy. Question placement can also affect the extent to which there is missing data on that question. Questions placed late in the survey are more likely to have missing data.
Now we will turn to specific examples of these different considerations. First, we’ll examine estimates of the uninsured derived from surveys conducted by the Census Bureau as well as those derived from state surveys. Then, we will examine estimates of dental visits across three national health surveys: the National Health Interview Survey, the National Health and Nutrition Examination Survey, and the Medical Expenditure Panel Survey.
Historically, the Census Bureau 's Annual Social and Economic Supplement to the Current Population Survey has been the most widely used source of health insurance estimates at both the national and state levels, particularly for estimating and monitoring the number and percentage of uninsured persons. Many states have conducted their own studies to derive estimates of the uninsured. In a study comparing current population survey estimates to state estimates, Kathleen Call and colleagues found that the state estimates of the uninsured were on average 23 percent lower than current population survey estimates. These were based on state point-in-time estimates. States’ full year uninsurance rate estimates exhibited an even greater deviation from the Census Bureau’s.
The differences in the Census Bureau’s Current Population Survey estimates and state estimates can be attributed to differences in survey questionnaire design, mode, and methodology. There have been a number of studies that have explored these differences and their implications for health insurance estimates. It is important to note that the Current Population Survey was not designed to provide reliable annual estimates at the state level.
More recently, the Census Bureau introduced the American Community Survey, which allows for more reliable estimates of household characteristics at the state level as well as smaller geographic areas. In 2008, the American Community Survey began including a question about health insurance coverage. There is ongoing research assessing the strengths and limitations of the American Community Survey with respect to generating estimates of health insurance coverage and how those estimates compare to the Current Population Survey and other surveys.
The next few slides provide a brief comparison of the Current Population Survey, American Community Survey, and state-level surveys on some of the key dimensions that were identified as being important considerations when comparing data across surveys.
The Current Population Survey is primarily a labor force survey, but includes supplemental questions that cover a range of socioeconomic and demographic topics, including income and health insurance coverage. The American Community Survey addresses a range of social, economic, and housing characteristics at smaller geographic levels. It has a single health insurance coverage question. State-level surveys are the most focused on generating health insurance coverage estimates specifically.
Although we haven't really discussed this yet, it is useful to know what data are available for what years, especially if there is interest in conducting longitudinal analyses. The Current Population Survey affords the greatest opportunity for conducting longitudinal analyses. The American Community Survey is relatively recent, especially with respect to health insurance coverage. Due to the resource requirements of conducting surveys, few states have ongoing surveys.
As noted previously, the Current Population Survey is designed to provide reliable national estimates, but it is not designed to provide reliable annual state estimates. The American Community Survey has a significantly larger sample and was designed to generate estimates at state and community levels. State surveys often have larger samples than the Current Population survey. But the samples may be smaller than those in the American Community Survey. However, state surveys have the advantage that they may target or oversample specific subpopulations of interest.
In terms of survey mode, the Current Population Survey relies on telephone interviews supplemented by in-person interviews. The American Community Survey is primarily a mail survey but it is supplemented with in-person and phone interviews. Due to cost considerations, most state-level surveys rely solely on telephone surveys and as a result do not enjoy response rates as high as the Current Population Survey and the American Community Survey.
Because health insurance coverage is not the primary focus of the Current Population Survey or the American Community Survey, questions about insurance status are placed late in the survey, whereas most state-level surveys ask key health insurance coverage questions early on.
There is variation in how information about health insurance coverage is elicited from respondents across the surveys. The Current Population Survey asks a series of questions for each major type of coverage, such as employment-based insurance, direct purchase, Medicare, Medicaid, and the Children's Health Insurance Program, or CHIP. These questions ask respondents to report if they or anyone in the household was covered by each type of insurance. States often create program names for their Medicaid and CHIP programs, such as Kid Care or Healthy Kids. The Current Population Survey attempts to include these state specific program names. The Current Population Survey also asks a verification question to confirm whether any household members were uninsured.
The American Community Survey asks a single question about health insurance coverage. A significant limitation of this single question is that several forms of government assistance are grouped together and there's no specific reference to CHIP. In addition, state specific program names are not used. These omissions may result in less accurate reporting. However, this question is asked for each person in the household separately.
This is in contrast to the Current Population Survey, where the questions ask the respondent to list the household members covered by each type of insurance. Unlike the Current Population Survey, though, the American Community Survey does not contain any follow-up verification to confirm uninsurance status.
Because the focus is on health insurance coverage, state-level surveys tend to have the most extensive questioning about insurance status. State-level surveys also have the advantage of being able to be more precise with state-specific program names and provide more program detail in the question wording to assist respondents in correctly identifying state programs, thereby improving reporting accuracy.
The key disadvantage of the Current Population Survey is its relatively long recall period. Respondents are asked to report in the spring of the current year about their own and their household members’ health insurance status for the prior calendar year. This long reference period results in recall difficulty that likely leads to over-reporting of uninsurance and under-reporting of coverage, especially for individuals who are covered for a short period of time during the prior calendar year. In contrast, the American Community Survey and state-level surveys typically ask respondents to report health insurance coverage at the time of the survey, which avoids the recall issues that are present in the Current Population Survey.
This last table summarizes the key advantages and disadvantages of each of the surveys. Among the Current Population Survey’s key advantages is its multi-mode approach that reaches no-phone and cell-phone-only households, which are frequently missed by state-level surveys. It has been fielded annually for more than three decades, which allows for trend analyses of coverage over time. Even though annual estimates may not be reliable for all states, 2-3 years of data can be pooled to examine variations in coverage between states and the Current Population Survey’s longevity makes it possible to examine these variations over time. The main disadvantages are the recall difficulty discussed earlier, and that it is not designed to provide reliable estimates for smaller geographic areas.
The American Community Survey is an encouraging development in national surveys of health insurance coverage because it does allow for state and local area analyses. It also uses a multi-mode approach. One of the key disadvantages is the use of a single health insurance question that is much less comprehensive and specific in asking about different types of coverage compared to the Current Population Survey and state-level surveys. Because the health insurance question was first included in 2008, currently there is limited ability to do longitudinal analyses and more evaluation needs to be conducted to determine its strengths and limitations in measuring health insurance coverage.
State-level surveys have the obvious advantage of being tailored to the topics and populations of interest to the state and, therefore, may be better suited to inform state policymaking and focus on subpopulations of interest. The key disadvantages are that the methodology varies across states, which makes it difficult to use these estimates for state-to-state comparisons. Due to the resource costs involved, state surveys frequently are conducted on a periodic basis which limits the options for conducting longitudinal analyses. Most states also rely on a telephone-only mode and, therefore, do not reach households without phones and typically have lower response rates than the Current Population Survey and the American Community Survey.
We will now turn to our second example, which examines estimates derived from national surveys of the percentage of adults reporting a dental visit during the prior year. For this example, we will review some of the key findings of a study published in 2002 in Health Services Research that illustrates the variation in estimates derived from the National Health Insurance Survey, the National Health and Nutrition Examination Survey, and the Medical Expenditure Panel Survey.
The estimates ranged from 45 percent for the Medical Expenditure Panel Survey to up to 67 percent of adults reporting a dental visit in the Third National Health and Nutrition Examination Survey. The differences in these estimates can be attributed to differences in survey question wording, response options, interviewer interpretation, and the reference period. The authors calculated estimates for the National Health and Nutrition Examination Survey using two different definitions of the prior year that varied by one day. As we will see shortly, they did so because the question was subject to interpretation. But this one day difference in interpretation affected the estimate by 14.5 percentage points. So, in this example we are going to take a close look at how variations in question wording and interpretation can influence survey findings.
Please note that the question wording on this and the following slides corresponds to the specific years analyzed in the study conducted by Macek and colleagues. Current wording can be obtained from each survey’s website. The 1993 National Health Interview Survey asked respondents to report the number of visits they made to a dentist during the past 12 months. So, the two key things to notice are the reference period and what the respondent was asked to report. The reference period was 12 months and the respondent was asked to report on the number of visits during that timeframe.
The Third National Health and Nutrition Examination Survey adopted a somewhat different approach. Instead of providing respondents with a reference period, the respondents were asked to report how long it had been since their last visit to a dentist or a dental provider. This allows for less precision on the part of the respondent and may invite responses such as “about a year ago” that requires interviewer interpretation.
Responses were converted into number of days in the analytic dataset from which the estimates were generated. As we saw earlier with the differences in estimates using a 364-day versus a 365-day definition, how one interprets the response for recording purposes and how the researcher defines one year in the analyses can make a substantial difference in the estimates generated.
The 1996 Medical Expenditure Panel Survey took yet another approach. Respondents were provided with a card that listed different types of dental providers and asked whether they had seen any of those dental provider types during a specific time frame. This survey uses several rounds of interviews with each household so the reference period is typically the preceding 3 to 5 months.
The differences in question wording likely contribute to the significantly higher reported rates in adult dental visits in the National Health Interview Survey and in the National Health and Nutrition Examination Survey. Specifically, the longer reference period in the National Health Interview Survey and lack of a reference period in the National Health and Nutrition Examination Survey are more likely to result in recall difficulty compared to the well-defined and shorter reference period in the Medical Expenditure Panel Survey, with a tendency for individuals to overestimate the number of visits when the recall period is more distant.
The National Health Interview Survey and the National Health and Nutrition Examination Survey questions may also be more susceptible to the social desirability phenomenon where people report based on their perception of social norms or expectations. That is, if the expectation is that one should have at least one dental checkup every year, people may be more likely to report having had a dental visit during the past year when given a twelve month reference period or an open-ended timeframe. When the year is divided into smaller segments and individuals are reporting on their experience for the immediate 3-4 months, these factors are less likely to influence responses. As a result, Macek and colleagues recommend the Medical Expenditure Panel Survey as the most reliable source of dental visit estimates.
Despite the significant differences in the overall estimates of dental visits, Macek and colleagues found that socio-demographic trends were consistent across the three national surveys. That is, if the interest is in examining disparities in dental visits, for example, between different racial and ethnic groups or by socioeconomic status, then similar results would be obtained from the three surveys.
It is also important to note that the authors observed that there were differences in the question format within the same survey, particularly the National Health and Nutrition Examination Survey, over time. And, since that study was conducted the dental visit questions have been revised in both the National Health Interview Survey and the National Health and Nutrition Examination Survey.
Therefore, when conducting longitudinal analyses it is critical to look at the specific question wording for each year in order to assess the comparability of the data. Fortunately, there are excellent resources to locate the specific question wording for these surveys over time. We will review some of these useful data sources at the end of the presentation.
We will now turn to the second main topic, which is using national survey data to conduct state and local analyses. As noted earlier, the resource requirements to conduct surveys are often substantial and limit the ability of states to have longitudinal data on a variety of health domains. Consequently, researchers and policy analysts often look to national surveys to meet their data needs.
When considering whether national survey data are appropriate for conducting state or local analyses, it is important to know first what geographic areas were included in the sampling frame and are reported in the analytic data files, and second, at what geographic level was the survey methodology designed to provide reliable estimates?
Many national surveys caution against using the data for analyses at the state and local levels. As the National Center for Health Statistics notes, even though there may be data available at smaller geographic levels for many national surveys, the survey methodology may only support reliable estimates nationally and for large regions.
Frequently, there's insufficient sample size for many states to have reliable single-year estimates. The sampling design is often a multi-stage clustered approach where sets of counties are sampled, but all counties within a state frequently are not included. However, because there is a significant need for state and local analyses, there have been increasing efforts over time to produce more reliable estimates at these smaller geographic levels. One approach is by introducing new surveys, such as the American Community Survey, and there are other national health surveys that have data available at the state level and for smaller geographic areas, such as the Behavioral Risk Factor Surveillance System, which may have data pertinent to the desired analyses. Another approach is using statistical techniques such as small area estimation.
For the national surveys that are not designed with state and local analyses in mind there are still some options for producing estimates of the state and sometimes, smaller geographic levels. Each of these options has important methodological considerations. One approach is to combine multiple years of data; this had been the recommended approach for examining state estimates of the uninsured using the Current Population Survey data. In doing so, it is important to verify that the variables of interest are measured consistently over time.
Another approach is to combine data from different surveys that share common questions. There are also a variety of statistical techniques, collectively referred to as small area estimation methods, that can be applied to national survey data to generate state-level estimates and sometimes estimates for smaller geographic areas. For example, researchers at the Agency for Healthcare Research and Quality have examined different methods for generating state-level estimates and have used small area estimation techniques to produce reports on dental expenditures and other types of healthcare expenditures in the ten largest states. In general, there are greater opportunities to obtain reliable estimates for the largest states, counties, and metropolitan areas that have greater sample sizes in the national surveys.
If you consider pursuing any of these strategies, it is important to be aware that geographic codes are often not provided in public-use files due to concerns about the risk of identifying individual respondents. Therefore, special requests must be made to obtain restricted use data files.
Given the limitations in using national survey data to conduct state and local analyses, it is natural to consider conducting your own local survey. Local surveys can be valuable because they are specifically designed to address local needs and policy issues. They also are more targeted to local populations and they may have more credibility with local stakeholders, including policymakers.
There two key potential pitfalls with designing and conducting local surveys. The first is not fully appreciating the complexities involved in designing and conducting a survey and then analyzing the results. Because most people can easily relate to surveys, there is sometimes a false sense of confidence that they are easy to do. However, surveys are resource intensive in terms of both the expertise and the funding required to carry out a valid and meaningful survey. The second key pitfall is lack of adequate resources in terms of both expertise and funding to get meaningful results.
Some of the complexities in conducting your own survey include identifying the appropriate target population; questionnaire design, which is the topic of another webinar in this series; sampling strategy; the targeted number of completed surveys needed to produce valid results; survey mode and the languages in which the survey is administered; survey administration -- that is, who will actually administer the survey; will the research team attempt to do this or will you work with the survey research center, which also is the topic of another webinar in this series; data collection and storage -- how will the survey responses be recorded and stored; quality monitoring, which can be conducted with respect to survey administration, data recording, and so forth. For example, at the Institute for Child Health Policy, we typically contract with a survey research center which has its own quality control procedures in place. But we also conduct our own monitoring of the interviews to ensure the survey is being carried out as intended and we also request data files early on to ensure that the data are being recorded correctly and in the format needed for our analyses. And last, but certainly not least, the appropriate methodology for data analysis, which includes calculating response rates, assessing non-response bias, weighting, and so forth.
When weighing whether or not to conduct your own survey, there are several key considerations that may help you make that decision. These include evaluating whether there are existing data sources that would meet your information needs. For example, in 2007 we conducted household surveys to estimate the number of uninsured children in Florida for the state. Following the subsequent economic downturn, the state wanted updated estimates but needed the results more quickly and at less cost than could be obtained through another survey. Therefore, we relied on secondary data sources and existing research about the relationship between changes in the unemployment rate and changes in children's health insurance coverage to generate updated estimates.
Another consideration is whether there are statistical techniques, along with the appropriate methodological expertise, that would allow you to extrapolate national or state data to the local area.
There may also be options to add questions to existing surveys. For example, each state conducts the Behavioral Risk Factor Surveillance System Survey. Although there are standardized core questions for this survey, states also have the option to add their own questions. You can check with your state coordinator for the survey to find out the options and process for requesting additional questions. Programs requesting new questions must typically identify funding to cover the cost of including these questions.
There may also be other ongoing state-level surveys conducted by state agencies or university research centers that may provide opportunities for adding questions that fit with the overall aims of the survey.
If you design your own survey, how well will that survey be able to meet the information needs? Surveys are resource intensive, so if the survey can only partially meet the information needs given the available resources, it may not be a worthwhile investment. The resource requirements in conducting surveys are not trivial. Therefore, it is critical at the outset to assess whether there are sufficient expertise and funding available for all phases of the survey to produce valid results. The Agency for Healthcare Research and Quality published a series on Monitoring the Health Care Safety Net. Joel Cantor contributed a chapter on Local Data Collection Strategies for Health Care Safety Net Assessment that provides additional considerations and resources for designing and conducting local healthcare surveys.
The last major topic we will address is linking state-level survey data with other data sources.
If there are state or local survey data from state agencies, local organizations, or surveys you have conducted, linking that data to other data sources can allow for richer analyses. For example, at the Institute for Child Health Policy, we commonly work with survey data from families whose children are enrolled in Medicaid and CHIP. These data can be linked to administrative enrollment and claims and encounter data which allows for more robust analyses than using either data source alone.
Even if you do not have data sources that can be linked at the individual level you may want to create linkages at other levels; for example, geographically by incorporating population or healthcare market characteristics at the county or census tract level. So, it is worthwhile to explore the different state-level data sources that may be available. The specific resources may vary by state but common sources of health data include state departments of health, the state agencies responsible for administering the Medicaid and CHIP programs, state agencies or research centers that generate state population estimates.
For oral health data specifically, state oral health coalitions can be a useful resource. Even if the coalition does not house the data, it will likely be able to direct you to the appropriate resources. Finally, university research centers often partner with state agencies in collecting and housing data that can be made available for public use.
When linking between different state-level data sources there are several important considerations. These include whether there are differences in the populations represented by the data sources and if the study’s target population is sufficiently represented in each data source. It also is important to identify for what time period each data source is available and whether the time frames are comparable.
Another key consideration relates to the specific variables contained in each data source. To link two data sources, there needs to be a common identifier. At the individual level, for example, is there a program identifier in each data source that uniquely identifies each person? For geographic linkages, how is the geographic area -- for example the county or the census tract -- represented in each data source? Are they standardized county codes, such as FIPS codes, local codes, or are they county names? If they are represented differently in the two data sources, what strategies will be used to align the two data sources and what quality checks will be put into place to ensure that the strategy resulted in the correct linkages?
In Florida, for example, Miami-Dade County may sometimes be listed as County Code 13 under Dade or it may be listed as County Code 43 under Miami-Dade, depending on the data source. So, even what seems like an easy linkage may not be as straightforward as it initially appears, and it is important to double check that the linkages are done correctly.
It is also important to verify how well filled each variable of interest is in each data source, and to assess the extent to which missing data may be a problem and how potential biases from missing data will be assessed and addressed. If possible, get a detailed data dictionary for each data source that specifies the variable names, provides a description of the variable, and explains how each variable is coded. For categorical data, a coding key is needed.
Even within the same data source, all variables of interest may not be available for the entire timeframe. On a related note, how a variable is measured and coded may change over time. For example, the coding for different racial and ethnic categories may change over time and how categories are coded may differ across the data sources.
Because there are often nuances to state and local data, it is useful to locate someone who routinely works with the data who can help you to better understand the data and identify measurement issues that could affect your analyses.
The National Health and Nutrition Examination Survey aptly sums up a key take away point in its analytic guidelines by advising data users to read all relevant documentation for the survey overall as well as for the specific data elements to be used in the analyses. While it is not the most exciting part of the analysis, following this recommendation will help to avoid some of the common pitfalls in comparing and linking survey data. In addition to survey documentation, there are numerous resources available to help researchers and policy analysts.
We will take a quick look at a couple of these resources.
An excellent starting point is the NIDCR/CDC Data Resource Center. For oral health researchers, one particularly nice aspect of this website is that it provides a summary of oral health questions from a wide range of health surveys organized by domain. So, for example, you could locate here the more current versions of the questions related to dental visits in the National Health Interview Survey and in the National Health and Nutrition Examination Survey. The catalog archive link in the left-hand navigation takes you to a searchable database that contains a wide range of local, state, national, and international surveys related to oral health.
In addition to the broad range of resources available through the Data Resource Center, it is also very useful to review the materials available on the specific survey websites. The National Center for Health Statistics is home to many of the national health surveys. Each survey has its own website with detailed documentation that includes the questionnaires and the survey methodology.
Several references are listed here for those who would like to pursue any of the topics presented in more depth.
[No audio this slide]
I’d like to thank the individuals who reviewed this presentation and provided helpful comments.
Thank you for your time and good luck with your survey analyses.