Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
1
An Introduction to Secondary
Data Analysis
What Are Secondary Data?
In the fields of epidemiology and
public health, the distinction between primary
and secondary data depends on the
relationship between the person or
research team who collected a data set and the person who is analyzing it. This
is an important concept because the same data set could be primary data in one
analysis and secondary data in another. If the data set in question was
collected by the researcher (or a team of which the researcher is a part) for
the specific purpose or analysis under consideration, it is primary data. If it was collected by
someone else for some other purpose, it is secondary
data. Of course, there will always be cases in which this distinction is
less clear, but it may be useful to conceptualize primary and secondary data by
considering two extreme cases. In the first, which is an example of primary data, a research team conceives
of and develops a research project, collects data designed to address specific
questions posed by the project, and performs and pub-lishes their own analyses
of the data they have collected. In this case, the people involved in analyzing
the data have some involvement in, or at least familiarity with, the research
design and data collection process, and the data were collected to answer the
questions examined in the analysis. In the second case, which is an example of secondary data, a researcher poses
questions that are addressed through analysis of data from the Behavioral Risk
Factor Surveillance System (BRFSS), a data set col-lected annually in the
United States through cooperation of the Centers for Disease Control and
Prevention and state health departments. In this case, the person performing
the analysis did not participate in either the
1
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
2
An
Introduction to Secondary Data Analysis
research design or data collection process, and the
data were not collected to answer specific research questions.
As an example of the same data
set serving as both primary and secondary data, consider the increasingly
common practice of one researcher performing an analysis of data collected by a
research team with whom he or she has no connection. This type of analysis is
facili-tated by the ease of sharing data stored electronically and the
concomi-tant creation of electronic data archives that allow access to
secondary users; some of these archives are discussed in Chapter 7. Such
analyses may serve a variety of purposes, such as addressing questions not
con-sidered in the original analysis or examining how a different analytic
approach might change the conclusions reached from the first analysis. In
either case, the same data set serves as primary
data for the original research team and secondary
data for the researcher performing the later analysis.
This book
deals primarily with secondary data in the sense of data sets that can be
obtained and analyzed in detail by the individual researcher. There is another
type of secondary data, again not mutually exclu-sive with the first, meaning
statistical information about some geo-graphic region or other entity. This
type of information is often useful to researchers: when you place your
research project in context by describ-ing the racial makeup or median house
value in the metropolitan area where you conduct your research, the data used
to compute those statis-tics were probably secondary data. Often these
statistics are computed on data collected by the federal government, and
Chapter 7 discusses several websites that were created specifically to permit
easy access to these types of statistics. In addition, many of the data sets
described in this book are accessible through an online interface that allows
the quick computation of basic statistics, without requiring the user to
download data and use a statistical program to analyze it. The availability of
such interfaces has been noted in the sections pertaining to each data set.
Most of the data sets discussed
in this volume contain either data col-lected through surveys or censuses, such
as the National Health Interview Survey and the U.S. Census, or administrative
records such as the medical claims records submitted to the Medicare system.
There are other types of secondary data, including diaries, videorecordings,
and transcripts of
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
3
Advantages
and Disadvantages of Secondary Data Analysis
interviews and focus groups: some of these are
included in sources dis-cussed in Chapter 7. Data such as interview transcripts
are often analyzed using qualitative data methods rather than the quantitative
techniques appropriate for most of the data sets discussed in this volume.
Secondary analysis of qualitative data is a topic unto itself and is not
discussed in this volume. The interested reader is referred to references such
as James and Sorenson (2000) and Heaton (2004).
Advantages and Disadvantages of Secondary Data
Analysis
The choice of primary or
secondary data need not be an either/or ques-tion. Most researchers in
epidemiology and public health will work with both types of data in the course
of their careers, and many research projects incorporate both types of data. A
more useful approach to this question is to focus on selecting data that are
appropriate to the research question being studied and the resources available
to the researcher; the latter include time, money, and personal expertise. In
this spirit, we offer a summary of the major advantages and disadvantages of
working with secondary, as opposed to primary, data.
The first major advantage of
working with secondary data is economy: because someone else has already
collected the data, the researcher does not have to devote resources to this
phase of research. Even if the sec-ondary data set must be purchased, the cost
is almost certainly lower than the expense of salaries, transportation, and so
forth that would be required to collect and process a similar data set from
scratch. There is also a savings of time. Because the data are already
collected, and frequently also cleaned and stored in electronic format, the
researcher can spend the bulk of his or her time analyzing the data. There is
also the influence of preference: secondary data analysis is an ideal focus for
researchers who prefer to spend their working hours thinking of and testing
hypotheses using existing data sets, rather than writing grants to finance the
data collection process and supervising student interviewers and data entry clerks.
The
second major advantage of using secondary data is the breadth of data
available. Few individual researchers would have the resources to collect data
from a representative sample of adults in every state in the
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
4
An
Introduction to Secondary Data Analysis
United States, let alone repeat
this data collection process every year, but the federal government conducts
numerous surveys on that scale. Data collected on a national basis are
particularly important in epidemiology and public health, fields that focus
primarily on the health of populations rather than of individuals. In addition,
some of the data sets discussed in Chapters 2 through 7 collect data using a
longitudinal design, and others are designed so certain questions are included
annually or at regular intervals, allowing researchers to examine the changes
in health status and health behaviors in the population over time.
The third
advantage in using secondary data is that often the data col-lection process is
informed by expertise and professionalism that may not available to smaller
research projects. For instance, many of the federal health surveys discussed
in this volume use a complex sample design and system of weighting that allows
the researcher to compute population-based estimates of health conditions and
behaviors. Although a local data collection project could conceivably use
similar techniques, more often a convenience sample, whose generalizability is
questionable, is used instead. To take another example, data collection for
many federal data sets is often performed by staff members who specialize in
that task and who may have years of experience working on a particular survey.
This is in contrast to many smaller research projects, in which data are
collected by students working at a part-time, temporary job.
One major disadvantage to using
secondary data is inherent in its nature: because the data were not collected
to answer your specific research questions, particular information that you
would like to have may not have been collected. Or it may not have been
collected in the geographic region you want to study, in the years you would
have chosen, or on the specific population that is the focus of your interest.
In any case, you can only work with the data that exist, not what you wish had
been collected. A related problem is that variables may have been defined or
categorized differently than you would have chosen: for instance, a data set
may have collected age information in categories rather than as a con-tinuous
variable, or race may have been defined as only White/Other. A third difficulty
is that data may have been collected but are not available to the secondary
researcher: for instance, address and phone number infor-mation for survey
respondents may have been recorded by the original
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
5
Locating
Appropriate Secondary Data
research team but will not be
released to secondary researchers for con-fidentiality reasons. If an analysis
incorporating geographic information was planned, such a restriction might make
the data set unusable. For these reasons, a secondary data set should be
examined carefully to con-firm that it includes the necessary data, that the
data are defined and coded in a manner that allows for the desired analysis,
and that the researcher will be allowed to access the data required.
A second
major disadvantage of using secondary data is that because the analyst did not
participate in the planning and execution of the data collection process, he or
she does not know exactly how it was done. More to the point, the analyst does
not know how well it was done and therefore how seriously the data are affected
by problems such as low response rate or respondent misunderstanding of
specific survey questions. Every data collection effort has its “dirty little
secrets” that may not invalidate the data but should be taken into account by
the analyst. If the analyst was not present during the data collection process,
he or she has to try to find this information through other means. Sometimes it
is readily available; for instance, many of the federal data sets have
extensive documentation of their data collection procedures, refusal rates, and
other technical information available on their websites or in published
reports. However, many other secondary data sets are not accompanied by this
type of information, and the analyst must learn to “read between the lines” and
consider what problems might have been encountered in the data collection
process.
Locating Appropriate Secondary Data
There is a vast quantity of
secondary data in epidemiology and public health that is available to the individual
researcher. However, the sheer quantity of data available, and the fact that
the data are collected and archived by many different governmental and private
entities, means that the process of locating appropriate secondary data is not
always straightforward. In fact, this book was written to ameliorate some of
the difficulties involved. There is no single process to be followed in every
case, but we offer two examples of the process of locating and analyzing
secondary data to address a specific research question or problem.
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
6
An
Introduction to Secondary Data Analysis
This
section might have been better titled “achieving a fit between your research
question and the data you choose to analyze” because it is often an iterative
process in which a research question is posed, potential data sets are
considered, the question is refined in terms of the data available, other
sources of data are considered, the question is refined again, and so on. The
most typical way to use secondary data for research is to begin with a research
question and seek a data set that will allow analysis of that question. An
alternative method is to begin by selecting from among the available secondary
data sets, and then formulating a research question that may be answered using
the data chosen. Although the first method conforms more to standard beliefs
about how research is done, the second approach is particularly useful in
classroom instruction, and both methods can produce quality research. If the
researcher begins with a question and then seeks out an appropriate data set,
the following generalized sequence of procedures may be useful:
1. Define the question you want to study; for instance, “How does the
experience of racism affect an individual’s health?”
2.
Specify the population you want
to study. Are you interested in chil-dren, adults, or people of all ages? What
races or ethnicities do you want to study? Do you want to analyze a national
sample or one con-fined to a smaller area? What is the range of years you would
consider (e.g., you may only be interested in data collected over the last 5
years)?
3. Specify what other variables you want to include in your analysis. In
this example, you might believe that it was important to have infor-mation
about the respondents’ race, Hispanic ethnicity, age, gender, income, and
educational level in order to include those factors in your analysis. If so,
you must confirm that the data you desire are contained in the data set that
you choose and that they are recorded in a manner that is useful to you. If you
are interested in comparing the experiences of Hispanic Blacks and non-Hispanic
Blacks, infor-mation about Hispanic ethnicity would need to be recorded in the
data set independently of information about race.
4. Specify what kind of data is most appropriate for your research
question: for instance, can it best be addressed through a national survey,
examination of hospital claims records, or transcriptions of
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
7
Locating
Appropriate Secondary Data
interviews? Also, specify if
there are any specific data collection tech-niques you believe are particularly
appropriate or inappropriate for your question. For instance, if you do not
believe people would answer questions about racism honestly in a personal
interview, you would not consider any data sets collected using that technique.
However, if you believe that a telephone survey would be the best way to
collect this information, you might begin your search by looking at surveys
that used this data collection method.
5. Create a list of data sets that include information related to your
research question and examine them to see if they meet your other requirements
(age range included, year of collection, etc.). This is where the interactive
process begins because you may have to revise either your question or your data
requirements, depending on the data that are available to you.
6. Once you have chosen your data set, examine the variables you intend to
use for the analysis of problems such as missing data or out-of-range values.
Also, read whatever information you can find about the data collection process,
data cleaning procedures, and so on in order to evaluate whether the data
quality is sufficient to meet your needs. If so, continue with the analysis; if
not, either devise a way to work around it (e.g., by imputing values for the
missing data) or choose another data set.
How do you generate the list
mentioned in step five? By any means necessary, as the saying goes. Consider
the data sets described in this book, search Medline to see what data sets
other researchers have used to address your topic, search the web portals
listed in Chapter 7, ask other researchers for suggestions, query relevant
email lists, and so forth.
If you
take the approach of beginning with a data set and crafting a research question
that can be addressed using it, the process is sim-ilar, but the order of
events is different. In this case, you would begin by looking at the variables
contained in the data set and considering how you might combine them to create
an interesting question. The process can begin with a germ of an idea, which
may reflect your per-sonal interests or a question that has arisen in your
work. For instance, you might be interested in how disability affects the
amount of physical
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
8
An
Introduction to Secondary Data Analysis
activity in which a person engages. You then need
to operationalize this question so it may be tested using the variables
available in the data set: how will you define disability, and how will you
define physical activity? At this point, a Medline search for related articles
would be in order, to see how others have addressed similar questions and
whether they have done so with the data set you will be using. This step will
help keep you from reinventing the wheel and will place your research in
context.
Alternatively,
you can begin by simply looking at the variables included in the data set to
see which of them interest you. For instance, if you were planning to work with
the BRFSS data from 2005, you might notice that eleven states included
questions on weight control procedures. You would then look at the actual
questions asked and confirm that the data were actually available. Information
to answer both questions can be found on the BRFSS website (http://www.cdc.gov/brfss).
This process should help you refine your focus so you can craft a research
question that can be answered using BRFSS 2005 data and that would add to our
understanding of public health. Because the BRFSS includes racial and ethnic
data, you might decide to look at racial and ethnic differences in weight
control practices. Or, taking advantage of the fact that BRFSS data are
identified by state of residence, you could plan to conduct a comparison of
weight control practices in different states. You could also plan a multilevel
analysis that combined information about state-level characteristics from the
U.S. Census (e.g., racial makeup or poverty level) with the individual-level
data available in the BRFSS. When you have selected the variables you will
include in your analysis, confirm that they are coded (or can be recoded by
you) in a manner that will support your intended analysis and that there are no
major data quality issues such as large quantities of missing data.
Questions to Ask About Any Secondary Data Set
Once you have located a secondary data set that you
think is appropriate for your analysis, you need to learn as much as you can
about why and how it was collected. In particular, you will want to answer the
following three questions:
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
9
Questions
to Ask About Any Secondary Data Set
1. What was
the original purpose for which the data were collected?
2. What kind
of data is it, and when and how were the data collected?
3. What cleaning and/or recoding procedures have been applied to the data?
Sources for this information
include the website of the agency or other entity responsible for collecting
and/or making the data available, pub-lished reports, research articles based
on the data, and personal com-munications with relevant individuals. For
instance, many of the federal agency websites include one or more contact
people who are available to answer questions about the data collected by that
agency, and a Medline search will often produce citations to reports and
articles discussing the procedures used to collect particular data sets.
The question of determining the
original purpose of the project that produced the data is important because its
influence may be present in other characteristics of the data, from the
population targeted to the specific wording of questions included in a survey.
Because you were not involved in planning phases for the project whose data you
will ana-lyze, you need this information in order to place the data in context.
To take an extreme case, you would certainly want to know if a research project
on the health effects of smoking was sponsored by a tobacco com-pany or by a
nonprofit dedicated to smoking prevention. You would also like to know if there
was any particular philosophy or model of health behavior that shaped the
project: for instance, was a smoking cessa-tion program structured using the
Transtheoretical Model? Knowledge of the core philosophical beliefs behind a
research project can illumi-nate the reasons for many choices made in the
planning and execution of the research and will be reflected in the end
product, the data you are proposing to analyze.
It is almost impossible to know
too much about the data collection process because it can influence the quality
of the data in many ways, some of them not obvious. To start with, you need to
know when the data were collected. A data set released in 2004 may have been
collected in the first 3 months of 2004 or over a 4-year period from 2000 to
2003. Second, you want to know the process by which the data were collected:
was it via telephone interviews, in-person interviews, abstraction of hospital
Cambridge
University Press
978-0-521-87001-6 -
Secondary Data Sources for Public Health: A Practical Guide Sarah Boslaugh
Excerpt
10 An Introduction to Secondary Data Analysis
records, or some other technique?
Third, you want to know the details of the data collection process. Questions
in this regard include who actu-ally did the data collection, how extensive was
their training, and how carefully were they supervised. If the data were
collected through chart review, what specific instructions were given to the
reviewers? If the data were collected through a survey, what was the response
rate? How many efforts were made to collect data from nonresponders? If data
were col-lected through a telephone survey, how were numbers selected? Was
there any attempt to correct for the bias introduced because households
with-out a telephone are not a random sample of all households? The issues of
survey data quality are the same whether the data set is primary or secondary.
For a thorough discussion of these issues, consult a reference such as de Vaus
(2002) or Bulmer, Sturgis, and Allum (2006).
The third
major question in working with any secondary data set is what was done to the
data after they were collected. For instance, almost all data sets include some
missing data. Were these data left as missing, or were values imputed, and if
so, how was the imputation done? Was any data cleaning done to remove
out-of-range values, and were those cases assigned missing values or was some
other procedure followed? Were certain combinations of answers considered
invalid, and if so, how were they treated? A famous example of this last type
of procedure was the decision in the 1990 U.S. Census to recode to the opposite
gender one member of a same-gender couple who declared themselves to be
married. If any recoding has been done, is it possible to restore the original
values? You also need to find out if data can be weighted, and if so, for what
aggregations the weighting allows the production of accurate estimates (e.g.,
at the national level alone or at both the national and the state levels).
Considerations Relating to Causal Inference
Causality and causal inference
are complex topics that can be touched on only briefly here. Issues surrounding
causal inference are discussed in greater detail by Rothman and Greenland
(1998) and Phillips and Goodman (2006). One of the first rules taught in a
basic statistics course is “association does not prove causation,” or because A
is
REVIEW
Di bidang epidemiologi dan
kesehatan masyarakat, perbedaan antara data primer dan sekunder tergantung pada
hubungan antara orang atau tim peneliti yang mengumpulkan satu set data dan
orang yang menganalisis itu. Ini adalah konsep yang penting karena sama kumpulan
data bisa data primer dalam satu analisis dan data sekunder lain. Jika data set
tersebut dikumpulkan oleh peneliti (atau tim yang peneliti merupakan bagian a)
untuk tujuan tertentu atau analisis yang sedang dipertimbangkan, itu adalah
data primer. Jika itu dikumpulkan oleh orang lain untuk tujuan lain, itu adalah
data sekunder. Tentu saja, akan selalu ada kasus di mana perbedaan ini kurang
jelas, tetapi mungkin berguna untuk konsep data primer dan sekunder dengan
mempertimbangkan dua kasus ekstrim. Pada bagian pertama, yang merupakan contoh
dari data primer, sebuah conceives tim peneliti dari dan mengembangkan proyek
penelitian, mengumpulkan data yang dirancang untuk menjawab pertanyaan spesifik
yang diajukan oleh proyek, dan melakukan dan pub-lishes analisis mereka sendiri
dari data yang mereka kumpulkan
Pemilihan data primer atau
sekunder tidak perlu menjadi baik / atau ques-tion. Sebagian besar peneliti
dalam epidemiologi dan kesehatan masyarakat akan bekerja dengan kedua jenis
data dalam perjalanan karir mereka, dan banyak proyek penelitian menggabungkan
kedua jenis data. Pendekatan yang lebih berguna untuk pertanyaan ini adalah
untuk fokus pada memilih data yang sesuai dengan pertanyaan penelitian yang
dipelajari dan sumber daya yang tersedia untuk peneliti; yang terakhir termasuk
waktu, uang, dan keahlian pribadi.
Peneliti dimulai dengan sebuah
pertanyaan dan kemudian mencari set data yang sesuai, urutan umum berikut
prosedur mungkin berguna:
1.
Tentukan pertanyaan yang ingin belajar; misalnya,
"Bagaimana pengalaman rasisme mempengaruhi kesehatan individu?"
2.
Tentukan populasi Anda inginkan
Apakah Anda tertarik penelitian tentang anak-anak, orang dewasa, atau orang-orang dari
segala usia? Apa ras atau etnis yang Anda ingin belajar? Apakah Anda ingin
menganalisis sampel nasional atau satu con-didenda untuk area yang lebih kecil?
Berapa kisaran tahun Anda akan mempertimbangkan (misalnya, Anda hanya mungkin tertarik
pada data yang dikumpulkan selama 5 tahun terakhir)?
3.
Tentukan apa variabel lain yang ingin Anda sertakan dalam
analisis Anda. Dalam contoh ini, Anda mungkin percaya bahwa itu adalah penting
untuk memiliki infor-masi tentang ras responden, etnis Hispanik, usia, jenis
kelamin, pendapatan, dan tingkat pendidikan dalam rangka untuk memasukkan
faktor-faktor dalam analisis Anda. Jika demikian, Anda harus mengkonfirmasi
bahwa data yang Anda inginkan yang terkandung dalam kumpulan data yang Anda
pilih dan bahwa mereka dicatat dengan cara yang berguna untuk Anda. Jika Anda
tertarik untuk membandingkan pengalaman Hispanik kulit hitam dan kulit hitam
non-Hispanik, infor-masi tentang etnis Hispanik akan perlu disimpan di data
diatur secara independen informasi tentang ras.
4. Tentukan
jenis data yang paling tepat untuk pertanyaan penelitian Anda: misalnya, bisa
itu terbaik diatasi melalui survei nasional, pemeriksaan rumah sakit klaim
catatan, atau transkripsi wawancara? Juga, tentukan apakah ada spesifik
pengumpulan data tech-teknik-Anda percaya sangat cocok atau tidak untuk
pertanyaan Anda. Misalnya, jika Anda tidak percaya orang akan menjawab
pertanyaan tentang rasisme jujur dalam
sebuah wawancara pribadi, Anda tidak akan mempertimbangkan set data yang
dikumpulkan menggunakan teknik itu. Namun, jika Anda percaya bahwa survei
telepon akan menjadi cara terbaik untuk mengumpulkan informasi ini, Anda
mungkin memulai pencarian Anda dengan melihat survei yang menggunakan metode
pengumpulan data ini.
5. Buat
daftar data set yang mencakup informasi terkait dengan pertanyaan penelitian
Anda dan memeriksa mereka untuk melihat apakah mereka memenuhi persyaratan lain
Anda (rentang usia termasuk, tahun koleksi, dll). Di sinilah proses interaktif
dimulai karena Anda mungkin harus merevisi baik pertanyaan atau persyaratan
data Anda, tergantung pada data yang tersedia untuk Anda.
6. Setelah
Anda memilih set data Anda, memeriksa variabel Anda berniat untuk menggunakan
untuk analisis masalah seperti data yang hilang atau out-of-range nilai. Juga,
membaca informasi apa pun yang Anda dapat menemukan tentang proses pengumpulan
data, prosedur pembersihan data, dan sebagainya dalam rangka untuk mengevaluasi
apakah kualitas data yang cukup untuk memenuhi kebutuhan Anda. Jika demikian,
lanjutkan dengan analisis; jika tidak, baik memikirkan cara untuk bekerja di
sekitar itu (misalnya, oleh imputing nilai untuk data yang hilang) atau pilih
yang lain kumpulan data.
Data sekunder yang baik adalah
memiliki tujuan yang tepat yang berdasarkan penelitan itu sendiri dan data yang
di kumpulkan sesuai dengan metode yang ada agar dalam pengumpulan data tidak
terjadi kesalahan atau out of range data yang mengakibatkan mengulang
pengumpulan data serta dapat memboroskan waktu dan biaya dalam penelitian.
Jurnal berasal:
