Research Questions to Results: Stats Analysis & Steps to Success in the Doctoral Capstone
Presented Wednesday October 18, 2017
Last updated 11/07/2017
Visual: Presentation opens with title slide “Research Questions to Results: Statistical Analysis & Steps to Success in the Doctoral Capstone.”
Audio: Instrumental music.
Visual: Slide changes to “Objectives” and lists the topics that will be discussed in the presentation.
Audio: Pat: So the three of us are going to be co-presenting this. There's really the objectives are to align the research questions to the variables in your statistical analysis. Then to go over power analysis and sample size calculations using the G*Power software. Building a data set using SPSS. Conducting an independent sample T-test and an analysis of variance using SPSS, and finally interpretation of SPSS output for both the independent sample T-test and the analysis of variance that we're going to run.
Visual: Slide changes to “Alignment of research question(s) to variables and statistical test for analysis” with Patrick Dunn, Ph.D., MS, and MBA.
Audio: Pat: So, again, my name is Dr. Patrick Dunn. And to add to my bio, I'm also a proud graduate of Walden University. I graduated with my Ph.D. in public health in 2015, so many of you that are students know that
I was on your side of the equation and thank you for your persistence and diligence in attending programs like this.
Visual: Slide changes to “Research Question No. 1” and includes an animated slide that fills in as Dr. Patrick Dunn discusses the topics.
Audio: Pat: So we're going to start by going back to the research question. You can't start just with the statistical analysis; you're kind of putting the cart before the horse. So always what you're going to do is first identify what research question am I trying to solve for. And so we're going to do two examples. In this example, is there a difference in cellphone usage for active versus inactive lifestyle? So to answer that question, we really have two major buckets of decisions. One is what variables do we have available to us and then, finally, what statistical analysis or statistical test will we use to answer that research question? So in this case, we have a dependent variable, average monthly minutes, and this is measured in minutes so it's a scale variable using SPSS. We also have an independent variable or a grouping variable, in this case, it's lifestyle and there are two groups, so there's active and inactive.
So just to see how many of you are actually paying attention, based on the research question in the two variables, does anybody want to guess what statistical test would be the best choice here? Go ahead and type responses into the chat. I'll wait a moment to see what kind of responses we get. Got a bunch of responses, looks like a bunch of people are typing so while you're typing so nobody gets put on the spot, we'll go to that. It would be the independent samples, T-test. That is the test that you would use if you have a scale variable, dependent variable with two choices so you're going to be comparing two groups feel.
Visual: Slide changes to “Research Question No. 2” and includes an animated slide that fills in as Dr. Patrick Dunn discusses the topics.
Audio: Pat: The next one is, is there a difference in the cost of phone service by age group? In this case, we have the average monthly bill and this is also -- this is in dollars so this is also a scale variable. And here we have a slightly different independent variable because we have more than two groups. We have age category, and there are four groups, okay, so those of you that know about T-tests know that you can't run a T-test with more than two groups. So, again, before I click the answer, if anybody wants to take a shot at what statistical test would be the best one here... And the answer to that would be the analysis of variance.
Visual: Slide changes to “Power Analysis for Sample Size Calculations using G*Power Software” presented by Matthew A. Jones, Ph.D.
Audio: Pat: Okay, I'm going to turn it over to Dr. Jones for the next part of the presentation.
Audio: Matt: All right, thanks, Pat. Hi, everybody, Matt Jones here, I'm going to talk about conducting a power analysis for our minimal sample size calculations for the research questions that Pat just went over. There are many different ways to calculate power -- well, many different mechanisms I should really say, software mechanisms. We're going to be focusing on the G*Power software here. G*Power is a free software download that you can get if you haven't seen it before, you just go ahead and Google or Bing, whatever your search preference is, G*Power, will probably be the first thing that comes up and is relatively easy to install so we're going to be using that for our example today.
Visual: Slide changes to “Research Question No. 1” and includes an animated slide that fills in as Dr. Jones discusses the topics.
Audio: Matt: So as you can see here, we've already gone over our first research question. Is there a difference in cellphone usage for active versus inactive lifestyle? If we were going to perform an analysis to answer this question, we really want to know something about power and power relates directly to our minimum sample size. So we want to know, what is the minimum sample size that I need to have enough power to be able to detect any effect that might exist within the data. And I'm going to talk a little bit more about that if we go ahead and walk through this here. So the output that you're going to see here is a copy and paste from output from G*Power. So when you run this, you'll be able to see something very similar.
So we already know going from what Pat told us that we'll do a verbal test under -- and under the family, this is under the family of T-tests. The specific type of test using G*Power language here is the difference between two independent -- that’s important, two independent means of two groups. The type of power analysis that we want to do is what we call and a priory and this will show up like this, a priory compute sample size given alpha, power and effect size. Those are the three pies; three key pieces that we really need to know to get a minimum sample size or alpha power and effect size. Now you can do a post-hock power analysis, I’m not really going to talk much at all about that today and there is a lot of literature that surrounds that, noting that it's not typically beneficial. You want to focus on the A priory, the beforehand, before you conduct the analysis, what is the minimum sample size. You always will be asked for tails, so this refers to whether this is going to be a one-tailed or a two-tailed test and whether that is the case really goes back to how you've stated your hypotheses.
So here we have a two-tailed test because we have -- let's assume, at least, a non-directional hypothesis, we just wants to know if there is a difference in cellphone usage between active and inactive, we're not stating a one-tailed hypothesis is that, I think, active is going to be more than inactive, it's just simply, I want to know whether there is a difference, either way, more or less between the groups. So that becomes a two-tailed test.
The next critical piece of information we need to enter is the effect size and there is a small D here and that refers to Cohen, the famous psychologist/statistician, Cohen's Q and .5 is often considered by Cohen's categorization, medium effect size. Typically if we know nothing about -- if we don’t have anything about the effect size, we haven’t run a pilot study, we don't have anything in the literature that mentions effect size, what we often want to do is go ahead and assume a medium effect size, that this is what we think is happening out in the population, this is what we think we're going to be able to detect at the medium level. If you play with that a little bit, if you go down and say, okay, I want to be able to see a small effect size, that's going to have a direct implication on your minimum sample size requirements. Your sample size requirements are going to go up; likewise, if you have a large effect size, they’re going to go down. So medium is sort of that middle of the road, if you will. Alpha error probability. So this is referring back to the P value and specifically your type 1 error, so what -- you know, when you enter this, we're thinking about the rejecting the null error and we want the probability of that to be fairly small.
Likewise, on the opposite side, we can think about type two error and power. So, power is one minus that and we can play with this number, as well, so right here again we have this -- what's often considered a conventional minimum threshold which is .80, so we want our studies to be powered at the .80 level. Now, you can go ahead and increase that threshold -- and I think in G*Power -- actually the default when it comes up is .95 which is much higher, so your probability of engaging in type two error in not rejecting a null when you should have or not being able to detect an effect out there in your data when one does exist, is lowered.
So what I would recommend and I think a lot of my colleagues recommend is just starting off with that .80. Obviously as you move this number higher, going up to .95, the default, again, that's going to have direct implications on your sample size. Your sample size requirements are going to increase. It's going to spit back a total sample size much larger if you have that higher power. So if you don't have G*Power and where once you do get G*Power and you want to play around with how the numbers work, I think that's a fantastic exercise, seeing if how you increased one, it affects sample size and decrease another, it affects sample size and so forth. Okay, the next piece of information we need is the allocation ratio. None over N2 and this is specific to our independent sample T-test.
When we say an allocation ratio for N1. N2 for one we're assuming both groups, the sample size is for both groups will be exactly equal. Now, in reality, that doesn't happen a lot unless you have a lot of tight controls and you're working with, say, in a psyche lab but, still, you can go ahead and assume that will be approximately -- and I really want to put emphasize on that word, approximately the same. So assuming there's one -- N one divided by N2 is one, sample all occasion is one for one between the two groups, our final -- the answer we're looking for, our final output that we need 128 unit, whatever the we're measuring, people, organizations, whatever your analysis, whatever the case is, we need 128 of those. So as you can see right here, we have 64 and 64, that equals 128 for the total sample size with an allocation ratio of one to one, there is 64 in each group, equally distributed. And that gives us the actual power of .80 and some change here, so we entered it as .80 and does some rounding that tells us -- the software tells us, okay, you need -- the power calculations it's doing behind the scenes for us,
you need a minimum of 128 -- I'm going to call them people as subjects for this, but whatever the that you're measuring, you need 128 people to be able to detect an effect and have sufficient power.
This becomes important because sometime you'll hear about studies that are quote, unquote, underpowered and we're cautious of those if we don't reject the null and they're underpowered, they have a small sample size and for those of you who are already in the dissertation or capstone phase, maybe your committee is already hitting you with questions about increasing your sample size, this is why we worry -- one of the reasons why we worry about sample size, if a study is under-powered and we didn't find an effect, we might be -- we might be engaging in a type 2 error, that is, we didn't reject a null when we should have. I mean, we don't know if that's the case because it’s under-powered. If we have enough sufficient power and we still fail to reject the null, then we can be a little bit more confident that we're not engaging in a type 2 error because that test is sufficiently powered. So that's how we did it for research question one.
Visual: Slide changes to “Research Question No. 2” and includes an animated slide that fills in as Dr. Jones discusses the topics.
Audio: Matt: Research question 2 is very similar except there are a few slightly different pieces. So, power analysis is conducted just a little differently for each specific statistical test, so when you're asking about minimum sample size related to power, you need to know what statistical test that you are using, so, for research question two, we have, is there a difference in the cost of phone service by age group? We already saw that this is going to be a one-way ANOVA, so moving down here in G*Power, under the test family of F-tests, wins the one-way NOVA gives a one -- statistic, and specifically in G*Power, this will be listed as fixed effects, omnibus, one way ANOVA. One way, one way. So sometimes just referred to as a global ANOVA, but this is what the language G-power uses. So, again, type of power analysis. A priory, so beforehand, computes required sample size, given power and effect size.
Again, this will be a drop-down menu for you in G*Power, this isn't any information that you have to hand enter by typing. From the same drop-down menu, tails. We want this to be a two-tailed test. We don't have any direction to our question or hypothesis, its non-directional and that’s perfectly fine. For effect size, if we had a Cohen's D for the independent sample T-test for the one way ANOVA, we have an F, again, assuming a medium effect that value is .25 and for these values for each individual effect size, you know, you can just go ahead and Google those out there, like what are the effect size for independent sample T-test, one way ANOVA, and there are some rough approximations. Going back to Cohen, he did a categorization some years ago and a lot of people use that it's in a lot of books so you can figure out what are the cut-offs between what's considered small, medium and large and they're all pretty close to each other. Again, same information as we entered before, our typical alpha error, probability of .05.
I already mentioned the power, you know, the minimum we want it to be .80 to let's start off with minimum because we're trying to achieve a minimum sample size. Unique to the power analysis in ANOVA, because we’re dealing with a factor as an independent variable that has groups, it has levels to it, right? So G*Power asks us, well, how many groups do you have for your factors? Well, as Pat mentioned, we have for, we know that, we have four groups. Okay, so the total power, total sample size, then, is going to be 180. Which gives me the actual power point again, .80 and some change. A priori, before we've run any analysis, so it's important to focus on this total this total sample size you see before an independent sample T-test, we talked about the allocation, so here's just telling us you need 180 across those four groups. Optimally from an analytical approach, we definitely would like to sequel distribution across those groups and I'm going to say approximately, again. The farther we move away from that, the more problems that can arise and we don't have time to talk about those here but when it's computing total sample size, the agreement is going to be proximate equal distribution across those groups. And I'm going to hand it over now to Zin about building this data set.
Visual: Slide changes to “Data Set building using SPSS” with Zin Htway, Ph.D., MBA, CT (ASCP, IAC).
Audio: Zin: Great, thank you, Matt. What I wanted to discuss today is about building a data set using SPSS. Based on research questions and the analysis presented by both Patrick and Matt. [Audio is difficult to hear]
Audio: Zin: There are two approaches you can do for building a data set. You can bring in your data first and then develop the variable screen or what I'm going to show you today is we're going to build actually the variable view first and then we're going to import in the actual data. Doesn't matter which direction you go, it's just that at the end, you need to come up with at the same point as where you have the data view shows all of your data and then in the variable view of SPSS, it actually has all of your variables laid out for you.
Visual: Slide changes to “SPSS data set [Variable View]” and includes an image of the software program SPSS. Slide is animated to circle the areas that Dr. Htway is discussing.
Audio: Zin: So beginning here, we actually see, this is the variable view of a new data set and we know that we’re in the variable view because in the lower left-hand corner of the screen, you'll see that there’s the variable view tab is highlighted in yellow. We begin which first column for our first variable which is going to be that cell name and row one, and what I suggest to student is try to select names that are easy. Much time I see data set where there is a lot of different code put in place and it's difficult to remember what the code actually mean, especially if you have 40 or 50 different variables. So it's always good to have an easy name to remember.
And in this case, what we're going to do is put in age category as we had once before. Once we put in age category, there are a number of columns here in the data set which you can go ahead and play with but the three that I'm going to focus on are actually name, the values column which is in the middle and then also the measure at the end. So we put in H category first and it comes up as a numeric variable. The default from SPSS is with an eight and the decimal places are two. You can of course adjust those as you need.
Now, we know that H category is an order natural measure so when we go to the far right under the order naturals column, we use the drop-down menu and change it to order natural. And then we'll take a look at the values column where currently it shows it's none which is highlighted in yellow. If we click on the right-hand side of the cell where it says "Values and none," we'll get a new menu that opens up; a new menu opens up in the center. One thing nice about SPSS is if the windows that open up are too small to read, you can use your mouse and grab the corner or the side and drag them to be a little bit larger. To go head and put in the value labels, into the values box, which is the top rectangular box, I put in the code of one and under the label I’m going to type in under 31.
I'm going to repeat that with the value of 2, and my age category is going to be 31 to 45. Repeat that with value 3 and the label is going to be 45 to 60, and repeat that with the value of 4, which is over 60. Each time you put in the value and then the values label, there will be a radial button that highlights that says "Add," you can go ahead and add. If you should happen to make a mistake, you can click on the actual values label and then select change or remove and then put in whatever adjustments you need. Once it's finished you click the okay button. So now we have our first variable put in place, so we can see that the name is age category, we know that it's numeric, I've changed the width down to four. The vowels label, we've put in the values label and we know that the measure is ordinal. -- Ordinal.
We'll repeat that with nominal and this time we have two groups that are active or not active, similar to what we have in terms of gender, there's no sequence to this, there's just two separate categories. Once again, the values column it's under none but we can go ahead and click on the right-hand side of that cell and when we do that, we'll put in our values. We're going to essentially dummy code zero equaling inactive and then one equaling active. Now, one thing that I want to emphasize here, when you put in these value labels, these labels will actually show up in the output of SPSS. If you leave these value labels blank, all that’s going to appear in SPSS are the actual dummy code numerical values which are going to be in this case zero and one or previously, it will be one, two, three or four. So it makes interpretation a bit difficult if you’re just looking at pure numbers. That's why I strongly emphasize and recommend that you fill in the planning’s here so that way -- blanks here, when you get to the output, it will be much easier to interpret. We'll click okay and then we've got our third variable that we put in, right, we got minutes, which is our average monthly minutes bill, that's a scale and then we also have the fourth variable which is bill, which is our average monthly bill, which is also a scale. Since both of these variables are scales, there’s no need to fill in the values column. We already know that minutes are -- fit the scale variable in terms of minutes and bill is a still variable in terms of dollars have the so now we switch over to the data view and you can see we're in data view because in the lower left-hand corner, the data view tab is highlighted in yellow. We can see our variable names and our measures are in the top row there. We've got age category, active, minutes and bill.
As I was saying earlier, I'm going to import in the actual data values now. I've got them on separate Xcel spread sheet. However, you have -- you keep your data or however you receive your data, it could be an excel spread sheet or a Word document, you can go head and copy and pays it in. I'm going to go ahead and do that now, so we fill in our age category. Next we have our active category. And we have our minutes. And then we have our bill.
Now, what's important -- I do emphasize, when you’re working on a large data set is to go ahead and continuously save the files as you move along. If you're working on a data set over a period of time, some of these data sets that we have students working with have thousands of cases, I strongly suggest that you save the data sets with unique file names. Add in the date, add in a letter after the date, and the reason being is that as you go through and start manipulating and working with your data sets, you'll have earlier iterations to go back to in the event that you should make an error somewhere along the way so you might realize that, oh, yesterday I actually put this one in and I shouldn't have, now I can go back to the previous version where I had it -- where I had all my data and then keep moving forward from there.
Visual: Slide changes to “Independent Samples t-Test and ANOVA using SPSS” with Patrick Dunn, Ph.D., MS, MBA.
Audio: Zin: Okay. At this point, I'm going to hand it over to Dr. Patrick Dunn to show you how to do the analysis.
Audio: Pat: Great, thanks, Zin.
Visual: Slide changes to “Independent samples t-test using SPSS” and includes an image of the software program SPSS. Slide is animated to circle the areas that Dr. Dunn is discussing.
Audio: Pat: Okay, so now we're ready to actually run the test. We're going to start with the independent samples T-test, using SPSS. So you're going to start in the analyze menu item up here on the top, so click on "Analyze." Now, take you to a menu configuration, so the first level is compare means, and then to the right of that, you'll see some additional choices so you're going to select the independent samples T-test. Note that there are on the types of T-tests and the one-way ANOVA is also in this group.
Once you select the independent sample T-test on the left you're going to have a list of your variables, and your dependent variable is going to be placed in the test variables. It can be a little confusing with SPSS sometimes because you don't always use the same naming conventions that we're using, so where it says “Test variables," that would be the same as the dependent variable. So you're going to put the monthly minutes in there, and then your independent variable is going to be in the what's called grouping variables box from the bottom, and note that once you put it in there, you're still not done, there are still some question marks and that's in this list, this dialogue that will come up. It will ask you, you have two groups, group one and group two and it will ask you what those groups are. So if you actually look behind the dialogue box, you can see -- and you can note from Zin’s presentation that when you built the data set, you coded those variables as zero and one, so group one is a zero, and group two is a one. So be sure to put those in there, in that grouping variable. And then you'll select "Continue."
And once you do that, you'll click "Okay," and you’ll get your SPSS output. If you've been using prior versions of SPSS, you'll note that the output, the default output now looks at least a little bit more like APA-formatted output, which is very nice, but the results are the same, even if you get one that looks a little bit more boxy with the output. There isn't the numbers -- it will be the same, the math has not changed. I'm not going to go over the result, we'll let Dr. Jones do that in a moment here but this is how we get the output stage for the independent sample T-test. Next we're going to do the same procedure for the one-way analysis of variance. And you're going to start in the same place, you're going to go to the analog -- I'm sorry she the analyze option, you're going to select "Compare means" and then the one on the bottom is the one-way ANOVA, you're going to select that. And your dialogue looks very similar so instead of test variables, now the top Bock, the dependent variable says dependent list.
Again, that would be your dependent variable so we’re going to put the average monthly minute in that box, and your factor is your independent variable, and that's our age categories. Once you've done that you've got a couple more selections to make. The boxes over on the far right is where we're going to be focused so you're going to -- just to go back to that, you're going to select the "post hoc" box there.
And you can see there are a number of post hoc tests available in this example, we'll choose the TUKEY post-hoc test and then you're also going to click on the options tab and you're going to select the homogeneity of variance test box. And once you've done that, you now have your SPSS output for the analysis of variance test.
Visual: Slide changes to “Results interpretation of Indepentedn Samples t-Test and ANOVA” with Matthew A. Jones, Ph.D.
Audio: Pat: I'm going to turn it to Dr. Jones now, he's going to explain how to interpret those results.
Audio: Matt: Thanks, Pat. So this is always the fun stuff.
Visual: Slide changes to “Independent samples t-test using SPSS” and includes and image of the data results of SPSS.
Audio: Matt: What does all the stuff that comes out of the test mean and how do you interpret that? So here we have our result from our independent sample T-test. So you can look up here right away, it gives us some descriptive statistics. I think these are always useful, so we have our two groups here that we're looking at, and our dependent variable, so average monthly minutes in a lifestyle inactive versus active. We have the actual sample size that was used to calculate that. And our descriptive around means and standard deviation in the average of the mean.
So moving down to the actual T-test that's down here, you'll see in independent samples T-test, you'll have two rows. Equal variances assumed and equal variances not assumed. And this really confuses a lot of people who are new to T-test and new to reading output because you need to know which row to read.
Now, more often than not, the results are going to be quite similar but sometimes there's just enough difference to make a really big difference in how you're interpreting your output, specifically are you going to reject the null or not? And you can kind of see that from this example here in just a second when we're going to talk about what the P value ended up being here. So the first thing to determine which row to read, we really need to pay attention to this first section of the output, and this is the Laverne’s test for quality of variances so that tills us read the bottom row or the top row and simply it's just asking a question, between the two groups, are the variances equal or not? So is the spread of data equal or not and it
tests the null hypothesis around that.
So here we want to look at our P value which is labeled as abbreviated as SIG, and SPSS and you can see that we have a fairly low P value, so .033, which is well below the conventional or arbitrary cut-off point for 05. We want to reject the null that the variances are equal; therefore, if we're rejecting the null that they're equal, we're assuming that variances are not equal. Okay, so that tells us we just need to follow this bottom row here, equal variances are not assumed. You see a critical T statistic for both, whether looking at the top or bottom is very similar and our associated significance, our P value detailed is very similar, .056 versus .053. But this is very close, very close to the threshold and there might be different circumstances fending on which row you read, it could make a big difference of whether your P value is going to be .048 or .053.
Which brings up the question for us today if we a priory, set our alpha level to be .050 and below to reject the null, then that means here we are above, although be it ever, ever, so slightly, we were just slightly above that. We still would not reject the null, so we would fail to reject the null because we said beforehand, I'm not going to reject the null in unless that pushes value is at .050 or below, and since it's just barely above, I have to stick to what I said and I'm going to fail to reject the null. Therefore, the null stand, the null that there is no difference between these two groups.
Now, sometimes you will see language like, you know, it was rapidly approaching statistical significance or it was very close to statistical significance, and that's your call whether you want to use that or not. This is just a really interesting example, you know, and I've run across this and you might at some date in the future where you're doing your analysis and it's so close that you'll have to make that judgment call of whether you want to reject the null or not. Either way, it's always important to report exact P values so that your audience understands when you say you're rejecting or failing to reject and you report that exact P value, they know that probability level that you were dealing with.
Visual: Slide changes to “One-way ANOVA using SPSS” and includes images of data output in SPSS.
Audio: Matt: For our one-way ANOVA, I know there is a lot of output here on the screen and it's kind of small but the main thing I really focus on here when looking at the ANOVA here, first of all, it gives us a test of the homogeneity of variances, we see Levine’s statistic there is a little bit low in the significance, it's not significant. And that becomes important because if we did have a significant omnibus ANOVA, we would want to make sure we're using the appropriate post-HOC test to do those pair-wise differences, to test those pair-wise differences across the four levels of our factor.
There are multiple tests for equal variances assumed, multiple tests for equal variances not assumed. But it becomes a little bit of a non-issue for us here in this output because right away, from this first piece of output where it just is labeled ANOVA, that's our omnibus ANOVA, that's the global ANOVA, and we have a P value going all the way over here to the S. G column, a P value of .094. Which is -- is certainly well above our .095, but, again, the conventional level, above the .05. So at least we know at that level, the .05 level, there are no differences in the data.
So, really, there's no need to do any post-HOC test unless you want to do a little bit of further investigation, there are times where some odd things can show up and it might be worth your time looking at the multiple comparisons just to do some double-checking but typically, if you see a global one-way ANOVA that is not significant, you’re done right there. And we can verify this here looking at our multiple comparisons and, again, we're just -- this tests all the pair-wise comparisons across our factors so the age group of under 31, to 31 to 45, the group of under 31 to 46 and 60, so, again, all possible pair-wise combinations, and you can see that we do not have -- just moving down the significance column here, you can see that we do not have a P value that falls below that .050 level. Therefore, there are no significant differences and when I say significant, I'm talking about statistically significant differences and the null in this case would stand. The null that there are no differences.
Visual: Slide changes to “Q&A Section”
Audio: Matt: Okay, I know we went through a lot of material here rather quickly so I know the three of us are here to open it up for questions. So I see the first question is how do you create a subgroup in SPSS? ZIN, you went over sort of the creating structure in data sets and I know this is something you like. Do you want to give that a try?
Audio: Zin: Sure, sure, I'll go ahead. To create the subgroups in SPSS, you basically have all of your cases and, say, for instance, we're looking at people and we have a thousand people and one of our subgroups is going to be ethnicity, then you want to create a variable for ethnicity and then fill in, in the values column, which... [No audio] Each of the dummy codes relates to which ethnicity. Once you have that built, you'll want to actually remember to put in the nominal measure, as well. Once you have that built, you can -- you'll have the subgroups there and in SPSS, if you wanted to, say, for instance, just look at African Americans, you can use the select cases function which is part of SPSS, select where ethnicity -- the code for ethnicity equals, say, 3 and by doing that, SPSS will isolate out just those cases where the ethnicity dummy code equals three and then you can work with your group that way. I hope that answers your question.
Audio: Matt: And I see Pat has already typed in the question for Evelyn but if I could add two cents, I'm not sure I add any value here to it but when you're doing mixed methods natural circumstances since mixed methods has a qualitative component and a quantitative component, then theoretically that you would need software packages for both a quantitative, like SPSS or R or another quantitative package and then for a -- you might need a qualitative package like NIVVO or maybe some hand-coding. So in mixed analysis, it's a case where you would need two packages.
Audio: Pat: There is a question about NVIVO. I don't believe that's available through Walden in the same manner SPSS is available.
Audio: Matt: Yeah, so we do not offer -- yeah, so this is Shawn saying we do not -- we do not offer availability to qualitative software package that I’m aware of, at least. Right now. But there are -- for those of you interested in qualitative research, there are a number of open source, that means free, qualitative software packages that are out there. I have had some of my own students use them in their dissertation research.
Audio: Pat: There is a question about G*Power, I believe G*Power is a free download. [No audio] Earlier in the chat that there was a price associated, unless something has changed, I do believe you can download G*Power for free. Just Google and put in G*Power and you should get the link. &
Audio: Matt: Yeah, Pat, last time I checked, it was for free, as well. I certainly downloaded it for free and I'm just on the web page right now and bullet point number one under "Download," is G*Power is free for everybody. That's good news.
I see there is a question of whether we have a list of open source qualitative packages. No, I do not think so but there really are not that many, to be honest, so if you just spend just a very little bit of time on the web, I think you would be able to find them. I'm thinking of one specifically. I think it's called hyper research or hypertext or something like that, which is a popular open source one.
Audio: Pat: And some of the qualitative packages also have student options. I know I used one -- I just got the student version so it was much less expensive than the full version, it was called max QDA.
Audio: Matt: Just a follow-up, I believe the package, the qualitative one I was talking about is called Hyper Research.
Audio: Zin: One thing I wanted to add about G*Power is -- I'm going to post a website here, web link, G*Power is really quite -- you can do quite a lot with it and sometimes it's difficult trying to find instructions on how to actually use G*Power for any particular statistical test. Today we used the sample of independent T-test and ANOVO which are fairly straight-forward but if you're looking to do ordinal logistic regression or perhaps even KI square, this website I posted the link to actually offers a -- a pretty good resource, you pick the test that you want and you scroll down and it's a good explanation for what values you need to put in and where because not all of them are always so straightforward and there are some -- I think there are some 60 different stay statistical tests that can be run through G*Power for calculating minimum sample size or achieve powered and achieved size and such.
Audio: Matt: Thanks, Zin. For power analysis can be very confusing so if you’re feeling a little bit at lost and this is your first exposure, or you've had multiple exposures to it, don't feel bad. It confuses the best of people. So just remember there are two key things we're focused on in hypothesis testing, type one and type two errors, so when we think about alpha, alpha is the probability of our type 1 error of testing correctly claim any statistical significance and then we have data which is the probability of the type two error. So in any -- we conclude there is no statistical analysis, the null remains and power is that one minus beta, beta being the probability of type two error, so that's that .80 value that we entered into G*Power.
Audio: Pat: A question about software for mixed methods research, my advice is it probably depends on the type of statistical or quantitative analysis that you're planning to use. Some statistics you really can't do by hand, you’re going to need a software program for them, whereas the majority of qualitative research you can do, you know, without a qualitative software package.
Visual: Slide changes to “Closing slide” and includes information about the Academic Skills Center, including email, and monthly updates with the Savvy Student Newsletter.
Audio: Shawna: Thank you for attending this presentation and I’ll close the session in about five minutes.