Research Methods in the Social Sciences

As a social science student, it is vitally important that you become an educated consumer of social science research, even if you never get to conduct an actual research study.

You need to have a basic understanding of the steps in the scientific process, a familiarity with commonly used social science research designs and methods, and a knowledge of procedures for collecting and analyzing social science data. 

This module provides very basic information about these topics. It addresses the following subjects:

· the scientific method

· qualitative and quantitative research

· data analysis and presentation

· discipline-specific research designs and methods

 The Scientific Method

The scientific method is a form of inquiry that was originally developed to answer questions in the natural sciences. The scientific method uses systematic observation and measurement to confirm or disconfirm hypotheses that most often are derived from underlying theories.

When a hypothesis is confirmed, it lends support to the underlying theory. When it is not confirmed, researchers must reformulate the theory and come up with other explanations. Scientists share their findings by presenting at conferences or publishing in peer-reviewed journals. Sharing results with the research world is an important part of the scientific method because science is an iterative activity.

As social scientists, we have a great challenge. Because we cannot exert complete control over human behavior, it is not possible to conduct true experiments. Real life cannot be captured in a lab, and human behavior cannot be explained by one or even two causes. This makes it difficult to determine "cause and effect." For example, if we are interested in knowing why poor people have compromised health outcomes, we would need to account for an impossibly large number of factors or variables such as race, ethnicity, genetic markers, income, education level, access to health care services, and living conditions.

Steps for Conducting a Research Study 

Figure 4.1 Steps for Conducting a Research Study

Source: UMUC PSYC 100.

The use of the scientific method allows researchers to develop explanations of social science phenomena. The steps of the research process are explained from a psychological perspective in this tutorial (Bradley, 2000).

Step 1: Make an Observation

The starting place for any research study is an observation (idea) regarding a topic of interest to you, the researcher. It may come from a conversation with a friend, a book or television program, or the conclusion of a related research project. All that is needed at this point is your interest in pursuing the topic further.

Let's take an example that we will use for all of the steps in the process. You observe that you are able to memorize items such as telephone numbers better when you eat candy. Because of this, you wonder if there is a connection between the primary component of candy (sugar) and memory.

Step 2: Propose a Hypothesis to Explain the Observation

Once you have your idea, the next step is to turn it into a statement called a hypothesis. A hypothesis is a testable prediction/belief of what will happen in some circumstance. Developing a good hypothesis requires an examination of the literature on that topic. Your research will help you to learn how other researchers have conceptualized and tested the hypothesis.

Example: You research the topic of memory enhancement and see that other researchers have looked at how certain herbs and a healthy diet can improve memory. However, no one* has examined the effect of sugar on memory. On the basis of your review of the literature, you formulate the following hypothesis:

"College students who ingest two grams of sugar before a memorization task will perform better than college students who do not ingest sugar."

(*Keep in mind that this is just an example and does not accurately reflect the research literature on memory enhancement.)

Step 3: Test the Hypothesis

In this step, you perform two important tasks:

1. You define all important terms and variables in your study.

As the researcher, you need to create an operational definition of all important terms in your study so that others can replicate it.

Example: "Sugar" and "memory" are terms that need to be defined. Sugar could be defined as the substance in white sugar cubes, while memory could be described as a score on a specific memorization test.

Since you control the amount of sugar ingested by subjects in this experiment, sugar is the independent variable. Scores on the memorization test are the dependent variable. Preexisting factors or characteristics can also be independent variables. For example, if some of the subjects in the study have Type II diabetes and others do not, diabetes could be an additional independent variable.

2. You decide on the best research method to test your hypothesis.

After you define the variables, you need to choose the best research method. The best method depends on your hypothesis. Psychologists typically choose from the five methods below. Click on each method for a definition and examples. Which would you choose for this study?

Figure 4.2 Social Science Research Methods

Source: UMUC PSYC 100.

Example: For our sugar study, we decide on the experimental method. We divide a class of college students into two groups. One group is given two grams of sugar cubes (the experimental group); the second group is given two grams of sugarless placebo cubes (the control group). After ingesting their cubes, the two groups have three minutes to memorize a list of nonsense syllables. After a five-minute break, both groups are tested on the number of nonsense syllables that they can recall. The number of correct syllables recalled represents their memory score.

Step 4: Analyze Data

After you have collected the data, you select the statistical procedures needed to prove or disprove your hypothesis. In the "sugar/memory" experiment, you compare the number of nonsense syllables that are correctly recalled by each group of subjects. The "Data Analysis" section of this module provides an introduction to statistical procedures that you might consider for this analysis.

Step 5: State Conclusions About Your Hypothesis Based on Data Analysis

The communication of research results is the final, important step in the overall process. Your findings are based on your analysis of the data collected. Researchers typically communicate their results through the publication of journal articles. They may also choose to present a poster or conduct a discussion at a professional meeting or research conference. Your findings may prompt others to conduct additional research that may result in new findings leading to additional research, and on and on.

Social Science Research 

Qualitative and quantitative research are two forms of scientific inquiry characterized by their differences in philosophy, methods, and results. 

Qualitative Research

In qualitative research, the establishment of trust and rapport with study participants is the foundation of the research. The researcher cannot obtain in-depth and accurate information about an individual's or group's experience without establishing rapport with the participants. Establishing rapport enables the researcher to gain access to important sources of data.

Qualitative researchers use various means of collecting information or data to help them understand the experience of study participants. One way of gathering data is through participant observation. The researcher may venture into the field and participate in rituals, ceremonies, or other activities of the individuals being studied. Another way of collecting data is by interviewing study participants. Interviews use open-ended questions to allow study participants to provide in-depth information about their experiences and understanding of people and events.

Ethnography is a common form of qualitative research. It is an in-depth study of an individual or group of people using observed, self-reported, and secondary data. One example of ethnography would be a study of the life story of a military veteran to better understand the impact of war on the veteran’s life and personal development. Data sources may include interviews, personal journals, and meaningful artifacts. Those sources are analyzed by examining important themes that emerge from those resources. They may also include linguistic, literary, and psychological analysis.

Quantitative Research

Quantitative research generates reliable, sample-based data that can be generalized to larger populations and used to establish cause-and-effect relationships. Data are expressed and recorded as quantities using a standard system of measurement. Quantities such as number, amount, time, frequency, or a host of others may be measured.

Quantitative research encompasses both correlational and experimental designs.

Correlational research examines relationships among variables. An example is the relationship between frequency/duration of exercise and diabetic health status (blood-sugar level). As the frequency/duration of exercise increases, the blood-sugar level decreases. This is an example of a negative correlation.

A positive correlation is one in which if one variable increases, then the other variable increases as well (similarly, if one variable decreases, the other variable decreases). An example of a positive correlation is the relationship between religiosity and perceived health status. The higher the level of religiosity, the higher the level of self-reported wellness or health.

A correlation or relationship between two variables doesn't mean that a change in one variable causes a change in the other. It simply means there is a relationship between two variables.

The correlation coefficient is a measure of the strength of the relationship between two variables. It can range from -1 to +1. The closer the coefficient is to +1, the stronger the positive association between variables. The closer the coefficient is to -1, the stronger the negative association between variables. A correlation coefficient of 0 means that there is no relationship between the variables.

Figure 4.3 Graphing Correlational Data: Perfect Positive Correlation

Figure 4.4 Graphing Correlational Data: Perfect Negative Correlation

Experimental research attempts to prove that a change in one variable causes a change in another variable. An experimental study manipulates an independent variable to examine its effect on an outcome (the dependent variable). Study participants are randomly assigned to an experimental or control group. The experimental group receives the intervention while the control group does not. Then the outcomes are measured and compared. One example of experimental research is the examination of the impact of a new health-promotion program for seniors on specific outcome or dependent variables such as food choice, sleep patterns, or physical activity. 

An understanding of certain basic statistical concepts such as variables, hypotheses, scales of measurement, populations, samples, and estimates is essential in the social sciences. This section of the module examines the meaning of these concepts by using an example.

Let's say that we want to look at how talkative people are in general. There is a basic assumption inherent in this interest—we assume that people vary in terms of their "talkativeness" (i.e., some talk more than others). Therefore, we consider "talkativeness" to be a variable. If we wanted to compare how talkative people are at work compared to how talkative they are at parties, then location (work vs. parties) would also be a variable. We now have two distinct variables, talkativeness and location.

In this case, location is our independent variable because we have some control over it (it may help to use the first letter of independent, i, to represent the phrase "I control"). Talkativeness is our dependent variable; we can measure it, but we cannot control it (here, you may think of the first letter of dependent, d, to represent "don't control").

Going further with our example, we will posit that people are more talkative at parties than at work. This would be our research hypothesis. The opposite of our research hypothesis is that people are either equally talkative at work and parties, or that they talk less at parties. These are called null hypotheses. In order to evaluate our research hypothesis, we need to figure out how to measure our two variables, location and talkativeness. Location is an easily-measured variable. You can only be in one place at a time (work vs. party)! You cannot be working a bit and partying a lot at the same time (presumably)! Therefore, the scale of measurement here is considered nominal. 

Talkativeness can be measured in many different ways. If we compare the talkativeness of some people to the talkativeness of others, then we have an ordinal scale of measurement. However, if we count the number of actual words spoken, we have a ratioscale of measurement.

We typically like to make statements about "people in general," but how can we do this when we are basing our results only on the friends who are at our party? On the other hand, is it feasible to invite all the people in the world to our party? Therefore, we are drawing conclusions based on the behaviors of a sample of the larger population. Nevertheless, by using the appropriate research methods, we can make a guess or estimate about talkativeness based on our sample situations.

As we move into the area of analyzing the data that we’ve collected, let’s look at a different example. Suppose we were to ask you, "How many movies have you seen per week in the past year?"

Let’s say you can only remember the number of movies that you saw during each of the last 10 weeks. This is a sample since you don’t have numbers for all 52 weeks. Suppose that the numerical values for the past 10 weeks of movies are:

Figure 4.5 Sample Frequency Data

Source: UMUC STAT 225.

Making Sense of Raw Data

Raw data such as numbers of movies watched per week per year only make sense when they are organized. There are two standard ways of making sense of distributions of raw data. The first method is aimed at counting the amount (or frequency) of data that is of a particular value or range of values. The numerical representation of all these individual frequencies of a distribution is called a frequency distribution. Translating the above raw data into a frequency distribution, we have:

Figure 4.6 Sample Frequency Chart

Source: UMUC STAT 225.

Central Tendency

Central tendency is one of the most basic concepts used to make sense of a data distribution. It is used to answer questions that contain words like usually, typically, and average; questions such as "How many movies do you see during an average week?" All of these words describe what we generally expect to happen. This is the core of the concept of central tendency.

Of course, nothing is that simple in statistics. We must have a few ways of describing central tendency in order to make it more complicated! There are three types of central tendency: the mean, median, and mode. Any distribution has all three types.

The Mean

The mean of a distribution is the mathematical "average." To obtain its specific value, you simply add all the scores and divide by n(the number of scores). The mean number of movies in our sample is 1.5. 

The Median

The median of a distribution is the value representing the middle point of that distribution, with half the scores falling above it and half below. To find the specific value, you simply use the formula, (n + 1)/2. For our movie example, n = 10. Therefore, the median would be the value associated with the (10 + 1)/2 = 5.5.

The Mode

The mode is simply the most frequently occurring score. Looking at our movie example, we see that in five out of the 10 weeks, you saw only one movie. One movie per week is the most frequently occurring score in this distribution. Therefore, the mode in this example is one. Sometimes two or more scores share the highest degree of frequency. In these cases, you have bimodal or multimodal distributions. While a distribution can have only one mean and median, it can have more than one mode.

Comparing the Mean, Median, and Mode

The mean and the median both take all the scores of a distribution into consideration when calculating central tendency. Furthermore, the mean takes into consideration each score's specific numerical value. The median uses the knowledge that a distribution has x number of scores to determine central tendency. The mode ignores all information except which score occurs most frequently. In the next section, we cover variability. We will see that different measures of variability are very similar to these three types of central tendency.

Let’s say we had different samples of how many movies you watched over the last year. Each sample recorded 10 weeks of movie watching, but no two samples had the identical set of 10 weeks in it. What we would probably see is similar mean values for all samples; in other words, the means would be more alike than either the medians or modes. You can also say that "the mean is the type of central tendency that is least subject to sampling variability." 

Graphing

The second method of making sense of a frequency distribution is graphing. Graphing takes the numerical summary of a distribution and provides a visual representation of it. Various types of graphical figures can be used, depending on the types of data and the way they are organized. Graphs offer the ability to step back and look at the big picture rather than getting stuck in the minutia of numerical values of a distribution. See the "Presenting Data" section for some examples of frequency graphs.

Variability

If we were to ask people how many movies they tend to see per month, the response might be "two, give or take one." The "two" is an estimate of central tendency and the "give or take one" is a guess about variability. This idea of variability is important because not all values in a distribution are the same. Values can differ a great deal or just a little. Variability gives us a sense of the extent to which scores are alike. It is another important characteristic of a distribution.

For example, another classmate says that "on average" she sees the same number of movies per week as you do. She says that the number of movies she has seen in the last 10 weeks, respectively, are:

Figure 4.7 Movies Per Week (Individual Data)

Source: UMUC STAT 225.

Organizing these data creates the following frequency distribution:

Figure 4.8 Frequency Distribution (Individual)

Source: UMUC STAT 225

The mean and median values for your classmate’s movie distribution are the same as those for your distribution (in the central tendency activity). However, by looking at them, you can see they differ in their variability. This section looks at standard ways used to measure this variability.

The Range

The range is a lot like the mode. Remember, the mode represented the most frequently occurring value. The range is similar in that it captures the highest and lowest scores of a distribution, neglecting the information contained in between. The range is useful to get a sense of how broad (or variable) is the score distribution. It is limited by its ability to describe the extremes, but not the internal variability of a distribution.

The Standard Deviation and Variance

What if we want to get a measure of how scores vary from some central tendency? Since the mean value is the one measure of central tendency that takes into consideration the value of every score, we use standard deviation and variance to perform this calculation. Standard deviation and variance are very similar concepts. The only difference is that variance is in squared units, and the standard deviation is not. 

Z Scores

We learned in the previous sections that the mean and standard deviation are significant characteristics of any distribution. Another significant characteristic is unit of measurement (e.g., meters vs. inches). Would you expect to get the same value for a person’s height in meters as in inches? Of course you wouldn't. But you would agree that they measured the same degree of height but just in different kinds of units.

Z scores solve that problem. They are a universal form of measurement. Transforming regular scores into z scores (otherwise known as standard scores) allows a comparison across different units of measurement. How can this be? Look at the example on height we just talked about. Transforming meters and inches into z scores allows an easy comparison of mean values (subtraction) and standard deviation (division). The two new z distributions should be identical (if the original measurements are precise). 

The Normal Curve

The normal or "bell curve" represents a normal distribution of data that always "behaves" in the same manner. A seemingly infinite number of datasets exhibit this pattern. Mathematical formulas (or functions) for a bell curve are identical with the exception of their mean values and standard deviations. 

If we were to take the means and standard deviations out of any normal distribution, we would be left with z scores. This is called "the z distribution." The standard normal (z) distribution allows us to find percentages, percentiles, and proportions associated with a single score or the area between scores.

Figure 4.9 Normal Distribution (Bell) Curve

Source: UMUC STAT 225.

Inferential Statistics

Inferential statistics provides us with the means to make statements about what is likely to be true about populations based upon information about a smaller segment of that population (sample).

For example, imagine that we want to see if a new type of medication helps allergy sufferers by decreasing the severity of their symptoms. It would be difficult, and very expensive, to test the drug on all the allergy sufferers in the world. So instead, we try it on a sample and see what happens for this smaller group.

Let’s say we find that, for the allergy sufferers in the sample, their symptoms are reduced by 50 percent. What we are really interested in is the whole population of allergy sufferers. So we use the information we got from the sample and guess what might be true for the population. In this case, we would estimate that on average, the symptoms of all allergy sufferers (a population) would be reduced by 50 percent if the medicine were used.

We use a sample statistic to provide an estimate of a population parameter. Inferential statistics allows us to evaluate whether our estimate is predictable or just due to chance. To help us decide whether the difference is due to chance, we use probability theory.

Probability and Statistical Significance

As scientists, when we conduct an experiment or make observations, we are hoping for definitive findings. The term "probability" (or likelihood) refers to how likely it is that an event will occur or that a statement is true. A probability is given as a percentage from 0 (0 percent chance of occurring) to 100 (100 percent chance of occurring).

If we get the same result in 100 percent of our experiments, we can be confident that our results are "real" (or true), but what conclusions can we draw if the same result occurs only 50 percent of the time? We may conclude that extraneous variables or conditions are affecting our results. If we get similar results when we "control" for these variables or conditions, we need to face the possibility that our findings are not real; they may be occurring by chance or coincidence. How do we calculate that likelihood? 

The term statistical significance refers to whether the probability of the difference between two results is real or due to chance alone. Statistical significance is expressed in terms of levels of confidence. We can say that the probability that a result is due to chance is less than 5 percent. We can also say that we are 95 percent confident that a research finding is real, and not due to chance. If we want to be even more conservative, we can say that the probability that the difference is due to chance is less than 1 percent, or that we are 99 percent confident that our result is not due to chance. These probability values are usually reported in research articles as p<.05 and p<.01, respectively.

Presenting data clearly is as important to the success of a research study as methods and statistical procedures. As you look at the following table and graphs, ask yourself why the researcher decided to present the data in this way.

Click on the individual figure number to check your response.

Sociologists seek to uncover the social factors that affect behavior. The goal of sociologists is to obtain data that test assumptions about the social world. The sociological perspective requires us to look beyond our common sense (which can be faulty), our experience (which can be limited), and our values (which are bound in time and culture) to gain new understandings of the social world.

Sociologists may use quantitative, qualitative, or mixed methods in their research. The methods used for a sociological study depend upon the research questions being asked.

Quantitative research in sociology employs methods such as surveys (developed to answer specific research questions) and secondary data analysis (research using existing data sets gathered for general purposes). A quantitative design attempts to amass information from large numbers of people. It often requires respondents to answer prepared (close-ended) questions. Close-ended questions are limiting in that they do not allow respondents to describe how they see or experience their world. 

Qualitative research employs some of the methods already discussed for other social sciences. Qualitative methods include interviews (asking people open-ended questions), content analysis (research to study content to uncover the explicit and implicit or hidden meanings), and participant observation (observation of, and involvement in, the social interaction patterns of groups). In qualitative studies, sociologists may ask subjects why they act in certain ways or what rules and assumptions govern their behavior. Questions in qualitative research instruments are more open-ended.

The following are limitations of qualitative methods:

· greater potential for bias (because researchers themselves can be influenced by their subjects)

· difficulty in generalizing findings to larger groups, given the small sample size in studies

What sets the social science disciplines apart from each other? Anthropologists, gerontologists, psychologists, and sociologists all select their research designs and methods based on clearly articulated research hypotheses and questions. Their designs and methods can be either qualitative, quantitative, or mixed. What truly sets the disciplines apart are the principles and theories that guide their research and interpretation of data.

For more information on those principles and theories, check out the Social Science Perspectives module.

References

Agarwal, P. & Searls, D. B. (2009, November). Can literature analysis identify innovation drivers in drug discovery? Nature Reviews Drug Discovery, 8, 865–878. doi:10.1038/nrd297.

Bianchin, J. N., Martendal, E., & Carasek, E. (2011, June 28). Determination of trace silver in water samples by online column preconcentration flame atomic absorption spectrometry using termite digestion product. Journal of Automated Methods and Management in Chemistry. Retrieved from http://openi.nlm.nih.gov/imgs/rescaled512/3142775_JAMMC2011-839365.008.png.

Bradley, M. E. (2000). The cyberlab for psychological research. Retrieved July 22, 2013, from http://faculty.frostburg.edu/mbradley/cyberlab.html

Hirschi, T., & Gottfredson, M. (1983). Age and the explanation of crime. American Journal of Sociology, 89(3), 552–584.

Nahin, R. L., Barnes, P. M., Stussman, B. J., and Bloom, B. (2009). Costs of complementary and alternative medicine (CAM) and frequency of visits to CAM practitioners: United States, 2007. National Health Statistics Reports No. 18. Hyattsville, MD: National Center for Health Statistics. Retrieved July 24, 2013, from http://nccam.nih.gov/about/plans/2011/introduction.htm.

National Institutes of Health, National Eye Institute. (n.d.). Congressional justification for FY 2007. Retrieved July 24, 2013, from http://www.nei.nih.gov/news/congressjust/cj2007.asp.

National Institutes of Health, Research Services Branch. (n.d.). Image processing and analysis in Java (plug-ins, Fraclac). Retrieved July 24, 2013, from http://rsbweb.nih.gov/ij/plugins/fraclac/FLHelp/Subs.htm National Kidney and Urologic Diseases Information Clearinghouse (NKUDIC). (2012, June). Kidney disease statistics for the United States (ESRD incident rate). Retrieved July 24, 2013, from http://kidney.niddk.nih.gov/kudiseases/pubs/kustats/images/Adjstd_Incident_Rates_ESRD.jpg