In the world of qualitative research, there are numerous designs from which to choose. Some, like autoethnography, focus directly on the researcher’s lived experiences and use creative, literary writing to present the study. Others take more of a postpositivist view and attempt to get as close to scientific research as possible in their data collection and analysis, but still maintain the spirit of qualitative research. This section will not include descriptions of all the qualitative designs available, but will focus on the five qualitative designs identified in Creswell (2009), ethnography, case study, phenomenology, narrative inquiry, and grounded theory.

Case studies are an in-depth and detailed exploration of a contemporary social phenomenon, event, or group (a case) in real-life context (Creswell, 2009; Patton, 2002; Yin, 2009). According to Yin (2009), case studies are most appropriate in answering how or why research questions. To answer these questions, case study researchers collect multiple forms of data including direct observations, participant interviews, relevant documents, and even quantitative data such as surveys, and triangulate the data through the analysis process (Yin, 2009). The findings of the analysis are reported in a narrative (or narratives for multiple cases) with thick, rich description to engage the reader and clearly communicate the results (Patton, 2002; Yin, 2009). Overall, the case study is used to bring the reader into the case experience to improve understanding (Patton, 2002) and “contribute to our knowledge of individual, group, organizational, social, political, and related phenomena” (Yin, 2009, p. 4).

Phenomenology is the study of human experience as it is lived (Moustakas, 1994). Phenomenology is conducted pre-theory and pre-understanding, meaning, phenomenologists attempt to rid themselves of preconceived notions surrounding the phenomena and enter into the study with an open and unbiased mind (Moustakas, 1994). To understand these experiences, researchers gather first-hand stories of the phenomena through long, open-ended interviews with participants in which they encourage participants to provide detailed descriptions of their experiences and perceptions (Moustakas, 1994). Throughout the process of data collection, the researcher engages in a process of continuous reflection of his/her own biases to ensure the experience or phenomenon is described through the eyes of the participant and not clouded by presuppositions of the researcher (Moustakas, 1994). The phenomenological study, then, is focused on description and the telling of stories from the understanding of the participants and focusing on the common themes of the participants’ experiences (Moustakas, 1994).

Like phenomenology, narrative inquiry is focused on stories of the research participants. Narrative inquiry begins with the researcher asking participants to tell their own stories through in-depth interviews or less structured conversations, or by using documents or artifacts as starters (Clandinin & Connelly, 2000). The field texts from these stories become the key data in the collection process (Clark & Connelly, 2000). During this process, Clandinin and Connelly (2000) described the researcher’s role as one of tension. The researcher records the stories of the participants while also building relationships with the participants, noting context, and recording his/her own stories while “continually asking questions of meaning and significance” (Clandinin & Connelly, 2000). In the end, a narrative inquiry is a literary work, using literary elements and rich descriptions which could also include autobiographical writings, varied forms of writing (i.e., poetry, letters) to discover meaning in an experience (Clandinin & Connelly, 1997).

Ethnography as a research methodology was born out of the field of anthropology when an anthropologist would live with and observe peoples of different cultures, often those in far off places, and report out the meaning they created from the experience (Patton, 2002; Wolcott, 1999). Although ethnography has evolved to include more systematic scientific methods and is no longer solely synonymous with travel, it is still characterized by extensive fieldwork and participation by the researcher (Moustakas, 1994; Wolcott, 1999). Because of the extensive fieldwork necessary, ethnography is usually conducted over a long period of time, and the researcher immerses himself/herself in the culture of the group being studied. In the field, the researcher conducts participant observations, interviews and converses with participants, and accesses and analyzes documents not readily available to those outside the group being studied such as diaries, archives, and written histories (Wolcott, 1999). According to Wolcott (1999), these techniques or methods are used to fulfill the purpose of ethnography:

To describe what the people in some particular place or status ordinarily do, and the meanings they ascribe to what they do, under ordinary or particular circumstances, presenting that description in a manner that draws attention to regularities that implicate cultural process. (p. 68)

Unlike the other methodologies described, grounded theory aims to do more than describe or explore a culture, case, or phenomenon, it seeks to generate a theory to better understand the meaning of an event or experience (Creswell, 2009; Strauss & Corbin, 1997). Grounded theory design does not begin with a theoretical framework, but begins with open questions that usually address gaps in literature (Moustakas, 1994). Then, the researcher collects data through qualitative methods such as in-depth interviews, observations, and document review while conducting data analysis simultaneously with data collection (Strauss & Corbin, 1997). As the researcher codes the research, he/she continues data collection and records not only new codes, but also his/her own thoughts during the research process (Moustakas, 1994). These codes are then sorted and compared to construct a theory (Strauss & Corbin, 1997).

As quantitative researchers, the actual results are compared with the chance model to determine if the results are statistically significant. The basic statistical question that we ask in quantitative design is, “Is what I found in my research significantly different from what I would expect to find by chance?” It is important to note that statistical significance does not mean “strong” or “important.”(In fact, you could have a very weak relationship that is still statistically significant.) Instead, statistical significance means that the results we found in our study are very likely NOT caused by chance. Researchers typically set significance levels (alpha levels) at .05 or .01. Therefore, a p-value less than either of those numbers (depending upon where you set your level of statistical significance) would indicate a finding that is statistically significant (or not likely caused by chance).

Every quantitative study has at least one independent variable (predictor variable or the presumed causal variable in a relationship) and at least one dependent variable (outcome or effect variable). The actual design that you choose for analysis of your data depends largely upon the types of variables that you are using (how you have operationalized and measured those variables).

A t-Test is a simple statistical procedure that you can use if one independent variable (IV) is categorical and one dependent variable (DV) is continuous. For example, if you wanted to know if there is a difference in scores between men and women on a specific achievement test, you would have two categories for your independent variable (men or women) and a continuous range of scores for the dependent variable (actual score on the test). In this study, your independent variable (IV) is a categorical variable with only two levels (or categories), and your dependent variable (DV) is a continuous variable. The appropriate test for analysis when you have one independent variable that has two and only two levels (or categories) and one dependent variable that is a continuous variable is an independent samples t-Test. The t-Test would be used to compare mean scores for men and for women on the achievement test by computing the ratio between the actual difference between scores and the difference that could be expected if findings were by chance. There are other types of t-Tests, for example paired samples tests, where we could seek to understand changes in performance after an intervention. For example, we could administer a pre-test to students before we taught a lesson, then teach the lesson, and then administer the same test again. A paired samples t-Test would help us to understand the influence of the lesson on student performance on the exam.

When the independent variable has more than two categories, the t-Test is no longer the appropriate statistical test. We still need a test that produces a critical ratio to detect departure from the chance model (Hoy & Adams, 2015). In this case, an F-Test using a statistical procedure called analysis of variance (ANOVA) is the appropriate statistical procedure.

For example, suppose you want to measure the effectiveness of three different types of diets on weight loss. Your independent variable would be the type of diet an individual was on (high protein/no carbs, counting calories, and a control group, for example). The dependent variable would be the number of pounds lost (or gained), so it would be a continuous variable. Thankfully, SPSS statistical software can run the calculation for you; however, the question is basically the same as it was for the t-Test: “Is there a statistically significant difference of weight lost (or gained) among the three groups?” In an F-Test, we are no longer comparing means (as we were with the t-test), now we are comparing variances across groups. A post hoc could tell us more about the influence of each diet on weight.

When both the independent and dependent variables are categorical, the Chi-Square test is the appropriate test. For example, suppose we want to know if gender is related to whether or not someone completes a difficult training course. In this case, our independent variable, gender, has two categories (men or women), and our dependent variable has two categories (completion or non-completion). Again, we are still answering the question of “Are these findings different than what we would expect by chance?” Two common types of Chi-Square tests are the Goodness of Fit and the Test of Independence.

So far, we have discussed types of analysis (the t-Test, F-Test and Chi-Square test) that are very helpful and can help us answer the question “Do my results differ from what I could expect by chance?” However, these tests do not tell us anything about the strength of the difference of our findings (effect size). If we want to know the strength of a relationship, or effect size, we turn to correlation or multiple regression. Each of these tests is described below.

Linear regression and correlation offer not only information about the departure from the chance model, they also offer important information about the strength of a relationship between variables. Correlations describe linear relations (Hoy & Adams, 2015) by quantifying the degree to which the variables are related, and this calculation is used when both the independent variable and the dependent variable are continuous. Correlation coefficients range in value from -1 to +1 (with -1 and +1 being equally strong relationships). The finding of -1 indicates a perfect negative relationship (meaning that when one variable goes up by one unit, the other goes down by one unit) and +1 indicates a perfect positive relationship (when one variable goes up one unit, the other goes up one unit; or when one variable goes down one unit, the other goes down one unit). Therefore, the closer the correlation coefficient is to “0,” the weaker the relationship. SPSS will provide us with correlation coefficients and levels of significance. For example, if our results are (r=.62, p<.05), the correlation is positive and the findings are statistically significant at the .05 level. That is, chance is unlikely to explain our findings (only 5 times in 100, the two variables would not be related). This finding also tells us about the strength of our relationship. The finding (r=.62) indicates what most researchers consider a moderately strong relationship. Researchers typically consider findings of up to r=.35 as weak relationships, r=.36-.65 as moderate relationships, and r=.65 and above as strong relationships. Obviously, these cut offs may differ slightly between researchers, and relationships are only considered “weak,” “moderate,” or “strong” if they are also statistically significant.

Linear regression allows us to model a simple relationship between a predictor (independent) variable and an outcome (dependent) variable. Through linear regression, we can understand and study relationships between two continuous variables when we want to model the relationship between the two variables by fitting a linear equation to observed data. In other words, linear regression allows us to fit a single line through a scatter plot of data points to understand the relationship between variables (using a statistical method called least squares). For example, if we wanted to understand the occurrence of skin cancer (Y) (as measured by the number of individuals diagnosed with melanoma in each state) and location of residence in states across the United States (X) (as measured by the latitude at the center of each state), we would hypothesize that harmful rays in states closer to the sun (X) would have higher incidence of diagnosed skin cancer (Y) (a linear, but not perfect, relationship). By plotting each of the data points on a graph where incidence of skin cancer was plotted on the y-axis and latitude was plotted on the x-axis, we could understand whether or not a linear relationship exists by drawing a line through the data points. Other examples include the prediction that, as height (X) increases, we would expect weight (Y) to increase or that, as the amount of sugar that a person eats (X) (as measured in ounces) increases, the number of cavities (Y) would also increase. The simple formula for linear regression is Y′ = bX + a (where bX = the slope of the line; a = y-intercept; Y = the dependent variable; X = the predictor or independent variable; b= the regression coefficient).

Multiple regression can tell us how much variation in the dependent variable (Y) is explained by multiple independent variables (Xi). Each independent variable in this equation will have a coefficient (called a regression coefficient or beta weight) (Hoy & Adams, 2015). Therefore, multiple regression will produce a beta weight for each independent variable (a value for each independent variable while controlling for the influence of all the other independent variables). For example, if we wanted to know the influence of family income (X0), number of years of parent education (X1), and number of children in a family (X2)on parent involvement (Y) (operationalized as number of hours parents spend involved with their children’s education, we would use multiple regression to compute the total influence of all three variables (income, years of parent education, and number of children in the family) on number of hours each parent spends being involved in educational activities. Additionally, we could understand the relative influence of each of those variables on the dependent variable (parent involvement) because beta weights would be produced for each of the independent variables. The formula is: Parent Involvement (# hours) = a + b1(family income) + b2 (years parent education) + b3 (# children in family)