Académique Documents
Professionnel Documents
Culture Documents
http://fmx.sagepub.com/
Published by:
http://www.sagepublications.com
Additional services and information for Field Methods can be found at:
Subscriptions: http://fmx.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://fmx.sagepub.com/content/22/2/154.refs.html
Qualitative Databases in
a Mixed Methods Study
of a Nontraditional
Graduate Education
Program
Abstract
A central issue for mixed methods research is for researchers to effectively
integrate (or mix) the quantitative and qualitative data in their studies. Despite
extensive methodological discussions about integration, researchers using
mixed methods approaches struggle with translating these discussions into
practice and often make inadequate use of integration in their studies. The
authors examined their integration practices as they applied three literature-
based strategies within the context of one mixed methods study about a
nontraditional graduate education program. From this examination, the authors
describe the processes, products, uses, and challenges that materialized as
they merged their quantitative and qualitative databases to develop a better
understanding of participants’ perceptions of their experience in the program.
1
University of Nebraska–Lincoln, Lincoln, NE
2
University of Texas at Dallas, Richardson, TX
Corresponding Author:
Vicki L. Plano Clark, University of Nebraska–Lincoln, Office of Qualitative and Mixed Methods
Research, 114 Teachers College Hall, Lincoln, NE 68588-0345
Email: vpc@unlserve.unl.edu
Keywords
mixed methods research, integration, concurrent design, quantitative and
qualitative data, data transformation
data sets are directly related or compared to each other. Creswell and Plano
Clark (2007) describe integration within a concurrent design as merging the
quantitative and qualitative data. The value of integration in concurrent ap-
proaches surpasses the mere summation of qualitative and quantitative
evidence; it is in the dynamic merging of the two forms of data that they
become greater than the sum of their parts. Therefore, the value of concurrent
mixed methods designs can be realized only if researchers apply effective
merging strategies in their practice.
This methodological article is aimed at increasing our understanding of
integration practices by examining three strategies for merging quantitative
and qualitative databases within one concurrent mixed methods study. To illus-
trate the strategies, we discuss examples from a mixed methods study about
a nontraditional science, technology, engineering, and math (STEM) gradu-
ate education program. The methodological discussion reports the processes
we utilized to implement each strategy, the research products that resulted,
the different uses we found for each strategy in response to our research ques-
tions, and the challenges that emerged in our practice.
Merging in a Discussion
The most straightforward strategy for merging quantitative and qualitative
data is to present and interpret the two sets of results in a conclusion section
of a manuscript (Creswell and Plano Clark 2007). Reviews of published
mixed methods studies have found this to be one of the most common ap-
proaches to achieving integration (Plano Clark 2005; Hanson et al. 2005;
Plano Clark et al. 2008). Some authors using this strategy present a paragraph
the alumni of the STEM graduate education program perceive the impact of
their participation? This question called for mixed methods because we needed
quantitative information to measure the perceived extent of the impact and
qualitative information to describe individual perceptions and experiences of
the impact. Thus, we included both types of data to develop a more complete
understanding of the participants’ perceptions.
Data Collection
The population for this study included all graduate students who had com-
pleted appointments in this nontraditional program before fall 2007. The primary
source of data collection was a Web-based questionnaire that included closed-
and open-ended items; that is, we collected the quantitative and qualitative
data concurrently. We chose to use an online questionnaire because (1) par-
ticipants were accustomed to reporting information about the program in this
format, (2) there were limited resources available for a longitudinal study
of program alumni, (3) quantitative and qualitative data would be readily
accessible, and (4) both kinds of data would be collected from all individu-
als in the sample.
The development of the questionnaire was informed by the findings of an
earlier study of the meaning of participation in the program (Buck et al. 2006)
as well as the stated goals of the GK–12 program. Topics included current
education and employment status, perceptions of program participation, beliefs
and knowledge related to being a scientist, beliefs and knowledge related to
K-12 education, and future professional plans. The questionnaire included three
items that included both closed- and open-ended parts, thirty-nine closed-
ended items, and five open-ended items (see examples in the appendix). We
invited individuals to participate via email and followed informed consent
procedures as approved by our institutional review board. The overall response
rate was 85.7% (N = 36). Although online questionnaires may produce poor
qualitative data, we felt that these individuals would take the time to provide
rich responses because they were invested in the program. In fact, the res
ponses to the open-ended items represented eighty pages of single-spaced text.
Therefore, although the data were not as rich as interview data, the resulting
database did allow us to explore individuals’ perceptions.
Merging in a Discussion
We found that merging results in a discussion was generally a straightforward
strategy to apply that was useful for many of our questions about specific topics.
Our implementation of this strategy followed five steps. First, we separately
analyzed the two sets of data to obtain separate sets of results and to explore
the available information. Second, we identified “overlap” topics within the
substantive content of the two sets of results. These overlap areas suggested
topics that would be fruitful for comparing different results. Third, we refined
the analyses in the overlap areas based on what we had learned from the two
data sets and as needed to facilitate comparisons. For example, we created a
quantitative subscale for communication skills based on its prominence in
the qualitative data. Fourth, we directly compared the quantitative and quali-
tative results for a substantive topic and assessed to what extent and in what
ways the data sets supported or illustrated each other. Finally, we wrote a para-
graph (such as illustrated in Figure 1) that (1) stated specific results from one
data set, followed by a corresponding result from the other data set and
(2) discussed our interpretations about how the result from one data set
corroborates, illustrates, or generalizes a result from the other data set. This
interpretation paragraph is the primary product from this integration strategy.
Different techniques. We utilized the merging in a discussion strategy for many
of our initial questions. From these questions, we used three distinct tech-
niques for merging by discussion: corroborating findings, developing a more
complete picture, and identifying divergence (see Table 1). The first tech-
nique was to develop a method by method comparison to corroborate results
about a specific topic or construct. This technique increased our confidence
in the validity and meaningfulness of conclusions based on separate results
from each database. For example, we wondered how the quantitative data
about the perceptions of the benefits derived from participating in the pro-
gram corroborated the qualitative themes about benefits. After analyzing the
two data sets, we compared the quantitative and qualitative results for specific
content areas, such as perceptions of improved communication skills. By
developing a discussion such as presented in Figure 1, we were able to both
describe a qualitative finding and indicate the extent to which it was supported
by the quantitative results on the same topic.
Merging results in a discussion also worked well to develop a more com-
plete picture by presenting two complementary sets of results about a topic.
This technique fit our online questionnaire data well, particularly when an
item included both a closed-ended and an open-ended part. For example,
participants rated their perception of the impact that participation had on their
progress toward completing their degree and described the impact. We asked,
how do the qualitative descriptions complement the quantitative ratings? By
presenting the combined results from both data sets in a discussion, we
developed a more complete understanding of the extent of the perceived
impact as well as the reasons for the ratings. For example, we were surprised
The theme of improved communication skills was a major emphasis for many
participants, most often the first topic they mentioned when prompted to discuss benefits that
came to mind. Three participants briefly mentioned improved written communication skills, such
as for grant writing. Most individuals, however, wrote about how their oral communication skills
had improved during the program experience. Some emphasized the improvement of general
public speaking skills, such as one who noted, “My public speaking skills have improved
QUAL dramatically.” Most participants who discussed this theme, however, specifically noted
findings improvements in their ability to communicate their content topic to different audiences. One
explained, “The interactions with teachers and children helped me to better articulate my
research.” Another added, “I gained valuable experience in communicating my research with
others.” Some participants noted that they had gained better appreciation of and skills for
communicating with individuals representing diverse backgrounds. As one participant wrote,
“Mostly, I learned a lot about communication... Sometimes it is just difficult to communicate, but
there are a lot of things that can be done to make it easier.” Improved communication skills also
emerged as a positive benefit from the quantitative results. Participants rated the items related to
QUAN oral communication (with groups and individuals) highly (M = 4.25, SD = .612), indicating that
results participants perceived these skills as improving because of the program. As noted in the
& qualitative findings, the quantitative results also indicate less of a benefit related specifically to
comparison written communication. The item about improved written communication had the lowest mean
value (M = 3.64, SD = .683) of all items asking about different improvements, with 17 participants
indicating that this skill had not changed as result of their participating in the program.
both quantitative and qualitative data, which has important implications for
data collection as well as data analysis. Although some scholars argue that
quantitative and qualitative data cannot represent the same phenomenon (Sale,
Lohfeld, and Brazil 2002), we successfully identified general “topics” of over-
lap and compared the results from the two data sets to examine complementary
facets of the participants’ perceptions on these topics. A key tension we ex-
perienced was between remaining true to the traditional standards of
analyzing the data sets and wanting to extract specific pieces of data for com-
parison. For example, on the topic of perceived benefits, we debated whether
to analyze our qualitative data using predetermined topic codes derived from
the closed-ended items to produce data that were easily comparable or to
conduct a separate thematic analysis to capture the richness of the qualitative
data. In the context of our exploratory purpose, we opted to conduct an initial
thematic analysis that began with open coding. This approach ensured that
we would find areas of convergence and also increased the likelihood of
uncovering areas of divergence, such as skills discussed in the qualitative
data that were not measured with the quantitative items. Another challenge
was deciding whether each comparison should be initiated by first examining
quantitative results or qualitative results. We found that being open to starting
with either data set encouraged us to think more deeply about meaningful
comparisons that could be made and the differing rhetorical effect that
resulted as we composed our discussion paragraphs.
QUAL
sample quotes. Last, we interpreted how the resulting matrix display uncovered
similarities and differences among the cases along the various dimensions.
Different techniques. We identified three techniques for using matrices to
merge the two data sets in response to our research questions (see Table 1).
These techniques differ in terms of how the grouping categories are derived
and the information displayed within the matrix cells. The simplest applica-
tion of this strategy was to explore differences in the qualitative findings
based on levels of a categorical variable (e.g., qualitative comments about
self-confidence by men vs. women).
We found two additional techniques for applying matrices in our study.
The second way we used a matrix was to qualitatively explore a statistically
significant difference in the quantitative results. We were interested to learn
how the perceptions of program benefits varied by certain characteristics, such
as disciplinary field. Through our quantitative analyses, we found significant
differences among disciplines (F = 5.21, p = .011), and univariate post hoc
tests indicated that the biologists’ perceptions differed significantly from those
of the physical scientists and engineers and the mathematicians. We developed
one matrix that displayed quantitative group means and qualitative themes
related to benefits as a function of discipline. The resulting matrix relates a
categorical variable, means for a continuous variable, frequency of occur-
rence of qualitative themes, and illustrative quotes (see Figure 2). This matrix
helped us understand the different ways that individuals benefited as well as
the disciplinary contexts for those perceived benefits (e.g., viewing one’s
discipline as competitive or placing a high value on teaching). A third tech-
nique emerged from situations where we examined results in terms of a grouping
variable that we derived from the qualitative data. We grouped participants
by types of negative experiences (program tensions vs. loss of disciplinary
experience) and then compared the groups in a matrix in terms of their quan-
titative scores.
Methodological challenges. Several challenges occurred as we merged our
results in matrices. First, this strategy requires identifying a grouping variable
from one data set that corresponds to meaningful differences in the other data
set. As we weighed different possibilities, we were reminded that the deci-
sion as to what is meaningful can be made only within the context of the study’s
purpose. This became important to emphasize because so many potential com-
parisons exist between the two data sets, and we found that without care, the
ease of applying this strategy could outweigh the use of the strategy to answer
specific questions. Although straightforward to apply for categorical vari-
ables, this strategy is more challenging when applied to continuous variables.
In one context, we found it helpful to convert participants’ scores into nor-
malized scores for grouping purposes. We note that both SPSS and MAXqda
have powerful capabilities for analyzing data by grouping variables. For exa
mple, by using the “attribute” feature of MAXqda, we were able to examine
our qualitative results in sets based on a quantitative variable, such as com-
paring the themes discussed by individuals representing each discipline. We
found, however, that summarizing the results in a matrix requires much tedious
copying and pasting among the different software programs.
New insights Identify the QUAL Develop and apply Establish evidence Conduct
call for information that will a rubric for for validity of new analyses and
creating a new serve as an indicator quantitizing the QUAN variable report QUAN
QUAN variable for the new variable QUAL information results
Quantitative Four themes that Rubric developed for Assessed inter-rater Tested to see if
data are emerged from the assigning a reliability and resolved the two groups
positively qualitative analysis “negative aspects” discrepancies; Tested (positive impact
skewed, but provide indicators of score to each hypothesis that “negative vs. very positive)
qualitative data negative aspects of the participant based on aspects” correlates differed in terms of
indicate that experience the number of negatively with the negative
there were unique statements “encourage others to aspects variable
Qualitative Information:
negative coded within the participate”
Emergent themes from
aspects about negative themes. New Result:
coded data: Program
individual’s Evidence: A significant
tensions, Negative
experience in New Variable : 100% agreement difference was
emotions, Degree took
the program Negative impacts (M achieved; Hypothesis found (Mann-
more time, and Loss of
= 2.82; SD = 1.65) supported (rS = –.483; Whitney U = 22.5,
disciplinary experience
p = .004) p = .004)
concern about the validity of the new variable, we employed strategies to estab-
lish evidence for its validity. These strategies included using multiple raters
during the scoring process, ensuring a high interrater agreement (at least 90%),
checking the characteristics of the new variable by evaluating its descriptive
statistics, and using convergent and discriminant validity procedures. Once
we were satisfied that the new variable provided useful information, we used
the new quantitized variable in statistical analyses with other quantitative vari-
ables and reported the results.
Different techniques. We used data transformation for two reasons: to develop
new variables and to construct better variables when variables that were needed
did not exist or were inadequate (see Table 1). For example, when designing
the questionnaire, we expected to find a range of scores in our quantitative
data, but instead the results were positively skewed across all items. Our
qualitative analyses, however, uncovered data about some negative experi-
ences with participating in the program. Therefore, we decided to transform
qualitative themes into a new quantitative variable (“negative impacts”) to
help us further understand the perceived impact of participation in the pro-
gram (see Figure 3). Because we did not have a theory to guide our quantitizing
of the negative experiences, we chose to count the number of statements
coded into the negative impacts subthemes (e.g., programmatic tensions and
loss of disciplinary experience) as an indication of the extent of a partici-
pant’s negative experience with the program. We proceeded to relate this new
quantitized variable to other quantitative variables in correlational analyses
for validation purposes. Finally, we established that there is a significant dif-
ference in the amount of negative comments made by those who rated the
overall program impact as positive versus very positive. Therefore, we con-
cluded that participants found program participation to be a positive experience
overall, but for some this experience was tempered by their perceptions of
corresponding negative impacts.
A second use for data transformation emerged when we realized that one of
our existing quantitative variables was inadequate. We included items that
asked for participants’ actions and intentions in working with K–12 education
in the past, present, and future. The data about participants’ actions in the past
indicated that many participants had misunderstood the phrasing of the item.
Therefore, we were uncertain about the validity of the scores representing
their future intentions. Caracelli and Greene (1993) describe a technique of
consolidating quantitative and qualitative information into a new variable.
This technique is similar to data transformation, with the exception that more
than one type of data is considered to generate the new variable. We created a
new variable for all participants based on their quantitative responses about
future intentions and their open-ended qualitative comments that discussed
their future intentions. We developed a rubric for examining the extent of evi-
dence for future involvement in K–12 education in both the quantitative and
the qualitative information and assigned a score (1–4) indicating the level of
evidence (none, little, moderate, extensive). We used this consolidated variable
in further analyses to identify variables that predict future involvement.
Methodological challenges. We found that data transformation helped us take
advantage of important insights that emerged from our analyses but that this
technique should be used with care. This strategy requires the researcher to
clearly identify the need for a transformed or consolidated variable and then
determine whether sufficient information is available on which to base the
new variable. This procedure also requires developing standard procedures for
transforming qualitative information into a quantitative score in a reliable and
valid way. One issue that emerged in our analysis was the need to set proce-
dures for handling cases that had missing qualitative data. Once the variable
was scored, we also needed to consider its characteristics to determine what
statistical tests were appropriate (e.g., using nonparametric statistics) and the
strategies that could be implemented to establish evidence for the reliability
and validity of the quantitized variable. Finally, we found that we must use cau-
tion when interpreting the statistical results from the new variable to keep from
overgeneralizing the meaning as it may have limited reliability because of the
unsystematic nature of the unstructured data on which it is based.
Discussion
Our goal for this article was to carefully examine our practice as we integrated
our quantitative and qualitative data within one concurrent mixed methods
study about individuals’ perceptions of the impact of their participation in a
Morse 1991; Creswell and Plano Clark 2007; Teddlie and Tashakkori 2009).
Although we describe the overall design of this study as a concurrent approach,
it is clear that our analytic processes and interpretations drew on the databases
in both concurrent and sequential ways as the analyses moved from being more
separate (when merging in a discussion) to more integrative (when merging
with a matrix or by data transformation). Therefore, the notion of the timing within
a mixed methods design may not be as simple as one overall classification.
In light of these reflections, we also recognize limitations in our project.
For one, we considered only strategies for merging and did not consider other
integration strategies such as those used in sequential mixed methods
approaches. Our overarching purpose was exploratory, and the extent of the
two databases constrained the types of analyses that we could implement. We
had a relatively small sample size for advanced statistical analyses and a rela-
tively large sample size for in-depth qualitative exploration, a common issue
in many mixed methods studies. Although respondents provided extensive
responses for the open-ended items, our qualitative database was also limited
by the fact that we were unable to probe for further details. This limits the
depth of our understanding of the participants’ perspectives and raises con-
cern about the adequacy of the database used for the data transformation
procedures in particular. In addition, it is possible that there was an interac-
tion between the closed- and open-ended items that biased the participants’
responses, as found in another study using a questionnaire for collecting both
data forms (Vitale, Armenakis, and Field 2008). Although the generalizabil-
ity and transferability of the results to other nontraditional programs are
limited, our process of merging two data sets may be independent of the
study’s specific data. So similar procedures, questions, and issues may occur
in studies using more sophisticated quantitative and qualitative analyses.
Despite its limitations, this study gave us a better and more nuanced
understanding of the perspectives of the alumni of this one nontraditional
program. Through merging the two sets of data, we were able to examine in
greater detail both positive and negative aspects of program participation, to
develop a more complete description of how participants rate the experiences
as well as the reasons behind these ratings, and to explore differences among
the participants’ perceptions and experiences. We also identified numerous
methodological issues related to our merging strategies that warrant further
examination such as handling missing qualitative information in data trans-
formation, developing procedures for validating transformed data, and
examining techniques for identifying and utilizing divergent information. In
addition, further work needs to examine integration strategies beyond those
classified as merging.
Appendix
Here are three exemplar items from the online questionnaire. A copy of the
instrument can be requested from the authors.
Acknowledgment
This article is based on a paper presented at the 2008 annual meeting of the American
Educational Research Association, New York. We thank our colleagues in the Office
Funding
The authors disclosed receipt of the following financial support for the research and/
or authorship of this article: This work was supported in part by the National Science
Foundation (DGE-0338202). The opinions, views, and conclusions expressed in this
article may not reflect those of the funding agency.
References
Bryman, A. 2006. Integrating quantitative and qualitative research: How is it done?
Qualitative Research 6 (1): 97–113.
Bryman, A. 2007. Barriers to integrating quantitative and qualitative research. Jour-
nal of Mixed Methods Research 1 (1): 8–22.
Buck, G. A., D. L. Leslie-Pelecky, Y. Lu, V. L. Plano Clark, and J. W. Creswell. 2006.
Self-definition of woman experiencing a nontraditional graduate fellowship pro-
gram. Journal of Research in Science Teaching 43 (8): 852–73.
Caracelli, V. J., and J. C. Greene. 1993. Data analysis strategies for mixed-method
evaluation designs. Educational Evaluation and Policy Analysis 15 (2):
195–207.
Creswell, J. W., and V. L. Plano Clark. 2007. Designing and conducting mixed meth-
ods research. Thousand Oaks, CA: Sage.
Creswell, J. W., V. L. Plano Clark, and A. Garrett. 2008. Methodological issues in con-
ducting mixed methods research designs. In Advances in mixed methods research:
Theories and applications, ed. M. M. Bergman, 66–83. London: Sage.
Creswell, J. W., V. L. Plano Clark, M. Gutmann, and W. Hanson. 2003. Advanced
mixed methods research designs. In Handbook of mixed methods in social and
behavioral research, ed. A. Tashakkori and C. Teddlie, 209–40. Thousand Oaks,
CA: Sage.
Fetters, M. D., T. Yoshioka, G. M. Greenberg, D. W. Gorenflo, and S. Yeo. 2007.
Advance consent in Japanese during prenatal care for epidural anesthesia during
childbirth. Journal of Mixed Methods Research 1 (4): 333–65.
Greene, J. C. 2007. Mixed methods in social inquiry. San Francisco: Jossey-Bass.
Greene, J. C. 2008. Is mixed methods social inquiry a distinctive methodology?
Journal of Mixed Methods Research 2 (1): 7–22.
Bios
Vicki L. Plano Clark is director of the Office of Qualitative and Mixed Methods
Research, a research service center at the University of Nebraska–Lincoln in Lincoln,
NE.