Just for Students, Random SLP

Research Methods for the Common Denominator: Part 2

I have no doubt that part 1 of this (very) mini-series was one of the most exciting blog posts you’ve read to date. In fact, you probably walked away from your computer feeling 1.0563 ounces smarter (yes, intelligence is obviously measured in ounces). As I sit here writing part 2 of my research methods informational post in my new reading glasses, I know I already feel just a tad bit more brilliant (yes, the reading glasses definitely help).

In this post, I’m going to impart upon you some facts about the statistics that you’re likely to find in research papers. To start off, there’s an important distinction to make between descriptive statistics and inferential statistics.

Descriptive Stats: describe or summarize data about your sample/group of subjects.

Inferential Stats: uses what you now know about your sample (since you just performed research) to infer about the population that your sample represents. So, if you were testing a new treatment method on a subject pool of 30 children with Down Syndrome, you would likely be inferring something about that treatment method for all children who are similar to your sample (i.e., kids with Down Syndrome).

Within descriptive statistics, you will want to consider a handful of different specific statistical measurements, starting with distribution. A frequency distribution generates a curve that shows you the frequency of responses/scores at different levels (e.g., different ages, different severities, etc.).

from: http://www.sciencedirect.com

The curve might come out looking like a normal curve, which is symmetrical, has a mean, median, and mode with the same value, and aligns with 68.2% of the population being within 1 standard deviation of the mean.

On the other hand, your curve might end up being a skew curve (positive or negative) or a kurtosis curve (leptokurtic, or platykurtic). Regardless of how your frequency distribution curve turns out, it’s important to understand that different curves imply different things about the effect of the independent variable on the dependent variable. In addition to considering the frequency distribution, you may also have information about the central tendency: mean (average), median (half the values are higher and half are lower), and mode (value that occurs most frequently).

Variability is a critical factor of descriptive statistics. The standard deviation tells us the average deviation of scores from the mean, and this range of variability might indicate that most scores were similar to one another, and therefore the mean can be confidently counted upon. If the scores are all over the place, the mean might not be very representative of the actual range of scores received by the sample.

Inferential statistics begins by testing a hypothesis. The alternative hypothesis hypothesizes some kind of effect of the independent variable on the dependent variable (e.g., X treatment will benefit Y population). Often this is what the researchers hope to find at the end of their study. The alternative hypothesis cannot actually be proven by statistical tests (although it can be supported); rather the null hypothesis (which says the independent variable will have no effect on the dependent variable) is rejected when the alternative hypothesis is supported. In order to reject the null hypothesis and support the alternative hypothesis, researchers use a cut-off value, or significance level (alpha), to decide the point at which the independent variable was effective and statistically significant. Typically, the significance level is <.05: if the observed p-value is <.05, the probability of this result occurring by chance is less than 5 in 100, and therefore can be attributed to a real effect of the independent variable. Although <.05 is the most common significance level, it’s actually just an arbitrary number and, at times, may lead to Type I or Type II errors.

Various characteristics can affect whether the results of a study are significant (i.e., p = <.05). A bigger effect size (more difference between the treatment group and the control group) typically supports significance. Less variability (aka a smaller standard deviation) also supports statistically significant results. Finally, a larger sample size is more likely to support statistical significance.

With all this being said, statistical significance is just one piece of the inferential statistics puzzle. Other statistical outcomes that look at testing differences and correlations must also be considered. However, I’d like to think your brain has worked hard enough for one day, so I’ll leave those explanations for another time and another place!

Just for Students, Random SLP

Research Methods For the Common Denominator: Part 1

Raise your hand if you consider yourself up-to-date on the advances in your specialty of our big, wide, wonderful field of speech and hearing sciences. Now take your hand and high-five yourself for being one of the mighty who take the “current research” part of the evidence-based practice (EBP) triangle seriously (remember the other points: clinical judgment and patient/client values?). While it’s not always easy to generate high-quality data about human behavioral sciences (people and their behaviors are…well…messy), smart people around the world are constantly publishing fascinating research about the very topics and issues that comprise our field. Now if you’re like me, stats and research scare the bageezus out of you. There are lots of big words and tiny numbers and percents and charts and vocabulary I thought only lived in GRE study books. BUT, never fear! This is the first in a 3-post installment to help break down some of the mystifying elements of research methods so that you can cruise through those peer-reviewed journal articles like a pro.

Let’s start with quantitative studies! Quantitative designs can be broken down into 2 types of studies: single-subject designs and group designs. Group studies can be further broken down into experimental, quasi-experimental, or observational studies. Of these different subtypes, experimental designs tend to be of the highest quality of evidence. In order for a research study to qualify as experimental, it must meet 3 critical criteria: (1) there must be a control group, (2) participants must be randomized into groups, and (3) there must be manipulation of the independent variable across groups.

Now what, you may ask, is an independent variable? Well, you can thank your lucky stars that I define variables right here, right now! The independent variable is the thing being manipulated as part of the study (e.g., the treatment). The dependent variable, on the other hand, is the outcome measure. Therefore, the purpose of a study is to determine how the independent variable (e.g., the proposed treatment) influences the dependent variable (e.g., improvement in some skill, decrease in some negative behavior, etc.). Easy, right? Now that we have the independent and dependent variables figured out, it’s important to recognize that there are likely to be extraneous variables that should be considered and controlled. These variables are those that confound our understanding of the impact of the independent variable on the dependent variable. They can be intrinsic (e.g., demographic characteristics of the participants or disorder characteristics that were not controlled in the study) or extrinsic (e.g., time of day for testing and setting for testing). The more these extraneous variables are controlled, the better the study will be and the more you can trust the outcomes.

Unlike group research designs, single-subject studies are often more feasible to perform while completing clinical work. They can contain a single subject or multiple subjects, but the key is that each subject serves as his/her own control. The goal is to begin by taking multiple baseline measurements of the participant’s performance. The treatment is then introduced, and multiple measurements are taken during this treatment phase. The treatment is then withdrawn and-you guessed it-more measurements are taken. You can continue the pattern of reintroducing treatment and recording how that impacts performance. The expectation is that performance increases when treatment is applied and decreases when it’s withheld. Although a basic version of a single-subject design might be an ABA…BA structure (where A represents a period of “no treatment” and B represents a period of “treatment”), this can be varied to include additional treatment types (e.g., ABAC…).

Here are a couple more terms to get you off on the right foot:

Prospective Studies: A research question is posed and then the study is completed to answer that particular question.

Retrospective Studies: The research question is asked after data was already collected (often for a different purpose), so you then go back and reanalyze data in order to answer your new question. In these studies, it’s too late to control the methods and extraneous variables…you just have to work with the data that was already collected.

Ok, that’s it! Did you survive? If the answer is yes, then double-high-five yourself because you now have a solid (ok, ok…basic) understanding of research designs and how to identify the design of the study you’re reading! Go ahead jump onto PubMED or your favorite database and test out your new research methods vocab knowledge! Installment 2 will delve into understanding outcome measures, so don’t be afraid to get pumped!