Latest / Professor Peterson’s Psychology Classroom / Unlocking Stats: From Questions to Conclusions

Transcript

Unlocking Stats: From Questions to Conclusions · Professor Peterson’s Psychology Classroom

• 0:05
  Welcome to Professor Peterson's Psychology Classroom,
• 0:09
  connecting the science of human behavior to everyday life.
• 0:13
  Here's Dr. Curtis Peterson.
• 0:17
  I want to start a discussion on kind of the broad ideas behind
• 0:21
  statistical and statistical procedures. There are 2
• 0:26
  kind of ways statistics are used. There's what's called applied
• 0:29
  statistics, and then there's
• 0:33
  the ones that we use in the scientific method. Now, they're both the
• 0:37
  same statistics. It's just they both are used for different
• 0:42
  reasons. For example, statistics in marketing is about trying
• 0:46
  to learn what the trends in that market are and where,
• 0:51
  you know, what age group should an advertiser be
• 0:55
  focusing on, gender, all of those types of things. That's marketing applied research.
• 1:00
  In science, we use it to see—
• 1:06
  to analyze some type of scientific question that
• 1:09
  we may have, such as what
• 1:15
  is the addictive mechanisms of opiate? All right, so that's
• 1:19
  a scientific type of question. It can also be an applied question. And we
• 1:23
  all use a different methodology to try and figure that
• 1:27
  out. So I want to start with,
• 1:31
  for our class, what I'm mostly going to focus on is more of the scientific
• 1:36
  application, mainly because what you'll find is in the applied
• 1:40
  setting, whether you're going into business, marketing, or even,
• 1:44
  you know, computer information systems, We use the same type
• 1:48
  of pattern of thought, meaning that we just don't go
• 1:52
  randomly collect a bunch of data and analyze
• 1:56
  it and see what happens. We use it— we, in all settings,
• 2:00
  we use a different type— we use a similar methodology so that
• 2:03
  our statistics remain strong and are consistent
• 2:08
  with what we're trying to figure out. Okay, so With
• 2:13
  that being said, I'm going to share my screen. What I want to start
• 2:17
  with to make things kind of clear is I
• 2:21
  want to start— actually, you know
• 2:24
  what? I'm going to do something different. I'm going to share my screen.
• 2:28
  Let's go ahead and go to the PowerPoint, and then I'll go to the video.
• 2:33
  Okay. All right. So the start
• 2:36
  of a statistic application always starts with a question.
• 2:41
  And this should, you know, in junior high, high school,
• 2:46
  elementary school, this should look familiar. It's quote
• 2:50
  unquote called the scientific method. This method is mostly used
• 2:54
  for analytical purposes and
• 2:57
  to really test a very specific question.
• 3:00
  Okay. So when we look at this, what we want to
• 3:04
  know is we always start with some type of question.
• 3:09
  And then we go to the literature, to any other sources
• 3:13
  of data, to see what we know about that question.
• 3:18
  And then the idea is, is that that leads to a very specific
• 3:23
  testable question, which was what we call a hypothesis.
• 3:27
  All right, and at this point, when we're— this is when we're starting
• 3:31
  to look at— so we develop a question.
• 3:35
  Right, right here. And we look
• 3:39
  and we see and we try to figure out what hasn't been answered about
• 3:42
  this. And that narrows it down to a very specific—
• 3:47
  and I'm going to say this— testable question. Okay,
• 3:51
  and this is what we call a hypothesis. And within a hypothesis,
• 3:55
  we're going to do one of 3 things.
• 3:59
  One of 3 things. Either we're going just to say the
• 4:04
  hypothesis is a discovery. We're going to go collect data and
• 4:09
  just describe a phenomenon. Okay? The second
• 4:12
  type of hypothesis is one where we're stating whether
• 4:17
  or not one variable predicts another
• 4:21
  variable. And then in the third
• 4:24
  one, we want to see if one variable is associated with
• 4:29
  one. And those are the hypotheses. And we put them in very measurable
• 4:34
  terms. All hypotheses should be something that you should clearly be
• 4:37
  able to measure. And that's the important part about this
• 4:41
  is it's very specific and measurable.
• 4:45
  When we're doing statistical analysis, there's something that we'll
• 4:49
  get to a little later called Type I and Type II error. And we'll get
• 4:53
  to that here in a few minutes, but basically what that is,
• 4:57
  is it's It's 2 types of error stating
• 5:01
  whether we rejected our hypotheses correct or failed to
• 5:05
  reject our hypotheses. And those are important
• 5:09
  things to do because there's some pre-statistical analyses
• 5:13
  we can do to reduce that error, and that's called a power analysis.
• 5:17
  But we'll get to that a little later. Again, this is just an overview.
• 5:22
  And so We'll get to
• 5:25
  the very specific things as we move along. All right,
• 5:29
  once we have our testable hypotheses, we come up with a way to
• 5:33
  test that hypothesis through some type— maybe it's a survey,
• 5:37
  observation, whatever it may be. And then we analyze
• 5:41
  it using any of the statistical methods, which we'll go over
• 5:45
  here in a little while. And then of course, in good scientific fashion,
• 5:49
  we share our results in the applied setting. We go
• 5:53
  to the the consumer, the people
• 5:57
  who asked us to analyze it and we share our
• 6:01
  results with them. So that's kind of how we go about it.
• 6:05
  Okay. So let's take a look.
• 6:08
  This is just another way of thinking about it is all
• 6:12
  of these. And we, again, we'll get into the very specifics about
• 6:15
  this. So if it's not very clear right now, we'll get there.
• 6:20
  All of this up here leads us to some type of question.
• 6:25
  And then we need to decide what's the best way to
• 6:28
  answer that question through what we call qualitative methods,
• 6:31
  which we will only get into a little bit here,
• 6:35
  quantitative methods, which is what this course is going to
• 6:39
  focus on, or something we call mixed methods, which is a mixed
• 6:43
  method of quantitative and qualitative research. Okay. And just
• 6:47
  to kind of put this— well, now let me go to the next slide.
• 6:51
  All right. So within quantitative and qualitative
• 6:54
  methods, what different types of analyses can we do?
• 6:58
  And again, this is just a quick overview.
• 7:03
  So as I mentioned yesterday, we have 2 types of
• 7:07
  statistical methods. And in the research literature, we call
• 7:10
  one descriptive. Which we went over. And the other one
• 7:14
  is called experimental, which in here we're going to call inferential.
• 7:19
  So you can— the experimental and inferential method
• 7:23
  can kind of be put in the same category.
• 7:28
  When we're talking about descriptive methods,
• 7:33
  what we're again trying to do— and this is the main one— we're trying to
• 7:36
  describe the data. We're trying to take each individual data point
• 7:40
  and make sense of that. We do that by looking at
• 7:44
  means, error,
• 7:49
  dispersion, and all of those things. So I mentioned,
• 7:52
  you know, averages and means,
• 7:56
  modes, those types of things, those things that tell us what the average
• 8:00
  or the central theme is.
• 8:03
  And then we have error, which we want to know how much error we
• 8:07
  think is in this data. And then we also
• 8:11
  have dispersion or variance, and variance is probably
• 8:14
  the more appropriate word. We want to know how much our data
• 8:19
  points vary from each other, and that's— so that gives us a good description.
• 8:24
  And then in the descriptive method, we have correlation. Okay,
• 8:28
  the basic definition of correlation is a test to see
• 8:32
  whether 2 measurable variables are associated
• 8:37
  with each other or not, meaning that as one score on one measure
• 8:41
  goes up, the score on the other measure goes up, or as
• 8:45
  a score on one measure goes up, the scores on the other measures
• 8:49
  go down.
• 8:54
  Regression is a method
• 8:58
  of trying to determine whether one variable predicts another
• 9:02
  variable. And maybe a very basic
• 9:06
  example is, does study time predict test scores?
• 9:10
  That's, that's a regression question. So what
• 9:14
  you do is you measure how much time someone studies,
• 9:19
  and then you measure the scores on some type of exam,
• 9:24
  and then you put them into what's called a regression equation.
• 9:29
  And what the regression equation will tell us
• 9:33
  is one, we can put— with the regression equation,
• 9:37
  we could put some random score in there and it will tell us the probability
• 9:40
  of that score. Okay, the probability of that prediction.
• 9:46
  But the other thing about regression is it tells us how
• 9:50
  much, how much the variance
• 9:54
  in one variable is predicted by the
• 9:58
  variance in the other variable. So we call this an effect
• 10:03
  size. Okay, and so in our example of testing,
• 10:08
  what we're trying to determine is how much of
• 10:12
  a test score can be explained by study
• 10:17
  time. And the number is put into a percentage.
• 10:21
  Usually, you know, it ranges most regression power analysis—
• 10:28
  not power analysis— effect
• 10:35
  size, there is. It usually runs between 25% to
• 10:38
  30%. In fact, I believe last time
• 10:42
  I looked at the test scores and amount
• 10:46
  of study time, I think the effect size was about
• 10:51
  30%, somewhere around there, which if you think about it, that's,
• 10:55
  that's if 30% of a test score is
• 10:58
  based upon your study time, that means
• 11:02
  the less study time you do, the more likely you'll get
• 11:06
  a C or below on that exam.
• 11:09
  So that's kind of basically how that works. All right.
• 11:13
  Factorial analysis, we will not worry about in this course.
• 11:18
  But what factorial analysis does is it takes, let's say,
• 11:23
  a whole bunch of questions that you've asked participants.
• 11:27
  And it determines which questions should be clustered
• 11:31
  together to create a category.
• 11:36
  And it's used, you know, you know,
• 11:39
  if you've ever taken one of those measures where it's going
• 11:43
  to tell you how outgoing you are, how intelligent
• 11:50
  you are. It gives you multiple types of scales.
• 11:54
  Those types of measures were based on a factor analysis where
• 11:58
  they took questions, see how they cluster together, and then tried
• 12:02
  to discover the themes behind those clusters of questions.
• 12:07
  But this is something that we might touch on but probably
• 12:12
  will not do within this course. Whoops.
• 12:18
  Okay. So when we're talking about experimental
• 12:22
  design, what we're trying to look at is things called like cause
• 12:25
  and effect. With any of these descriptive methods,
• 12:29
  even with regression where we're trying to determine whether one variable predicts another,
• 12:35
  we can't state cause and effect because
• 12:40
  there can always be what's called the third variable problem, meaning that,
• 12:45
  you know, if we look at— I'll go back to the study time
• 12:49
  and the test score. So yeah,
• 12:52
  we came up with this number of 30%, and we
• 12:56
  know we can put scores in there and get fairly close to what the score
• 13:00
  would be. But what
• 13:05
  if it's not study time? What if it's
• 13:10
  people who study more tend to have
• 13:15
  a better situational environment to study in. Okay, maybe individuals
• 13:19
  who study less, they're dealing with a lot of chaos in their life,
• 13:23
  they're dealing with family issues,
• 13:25
  illnesses, anything you can think of. All of those are third variable
• 13:30
  problems that then could explain the the
• 13:35
  analysis we did. Okay?
• 13:38
  Probably in our modern time,
• 13:42
  one of the biggest blunders in correlation and regression
• 13:46
  that has been done is in the '70s
• 13:50
  and '80s, there was a famous study
• 13:53
  called the Minnesota Twin Studies, which tried
• 13:58
  to look at the association between
• 14:02
  IQ score and genetics.
• 14:06
  And the way that they do this is that
• 14:12
  you take monozygotic twins, which are
• 14:17
  twins who are born with the same— from the same egg
• 14:21
  and sperm. So they're genetically identical.
• 14:27
  And then you take dizygotic twins, which are twins who are born
• 14:31
  from 2 separately fertilized eggs. So they're no different
• 14:35
  than fraternal twins. They, they have the same genetic
• 14:39
  variation between any sibling. And then you look at those
• 14:43
  who are separated at birth.
• 14:46
  And you give them certain measures. And the
• 14:49
  idea is, is that for the monozygotic twins,
• 14:53
  if genetics really matters,
• 14:57
  then whatever you're measuring, they should be associated—
• 15:00
  those scores between those monozygotic twins should be
• 15:06
  correlated with each other. They should be associated with each other. And sure
• 15:10
  enough, in this Minnesota Twin Studies, they found
• 15:14
  a strong association, meaning for IQ experiments,
• 15:19
  strong associations is, you know,
• 15:23
  well, we'll get into that later, but just know that it was really significant.
• 15:28
  And so here we have had for the last 30
• 15:32
  years now this study that says there's a strong association
• 15:37
  between genetics and a person's
• 15:40
  intelligence score, or IQ score.
• 15:45
  And it's gotten to the point where we've started to talk
• 15:49
  about the genetic cause of IQ.
• 15:52
  Okay, well, just recently, in fact, this has happened
• 15:57
  in the last couple months, they went back and they reanalyzed that
• 16:01
  data. But the thing that they found out in the original
• 16:05
  data is they didn't run the same statistical test
• 16:10
  for the dizygotic twins as they did with the monozygotic
• 16:15
  twins. They, they, they used monozygotics as
• 16:18
  what we call a control But the similarities
• 16:22
  between mono— dizygotic twins was never tested
• 16:27
  until most recently. And sure enough,
• 16:30
  we have found that there is just as strong association
• 16:35
  between dizygotic twins and their IQ scores.
• 16:39
  And it's just as strong as the correlation between monozygotic
• 16:44
  twins and IQ scores. So we're
• 16:48
  kind of back to the drawing board because we have the third variable problem.
• 16:52
  There really is— there is an association between twins,
• 16:57
  but it cannot be at this point attributed
• 17:00
  just to genetics. It can't. So the association
• 17:05
  between IQ and genetics has been
• 17:09
  greatly weakened, and it's— and now we have to go back to the
• 17:13
  drawing board and really look at this very specific
• 17:16
  problem even closer. I just wanted to bring up that
• 17:20
  example to state that we can never use correlation regression to
• 17:25
  determine cause and effect. That's a question that can't—
• 17:29
  that really needs
• 17:32
  to be taken home really well.
• 17:38
  So it's in the experimental model that we look for
• 17:42
  cause an effect.
• 17:45
  Because what we're going to do in the experimental
• 17:49
  model is we are going to take something called
• 17:54
  an independent variable and
• 17:58
  we're going to see how it influences something
• 18:02
  we call a dependent variable.
• 18:08
  So let me explain what we're going to do with these tests.
• 18:12
  We're going to take an independent variable. All right,
• 18:16
  let's say— let's keep the same example
• 18:20
  we have been having and we're going to do study time,
• 18:25
  this SD, and we're going to put— and we're going
• 18:29
  to use ES for exam scores.
• 18:32
  And what my hypothesis is stating is my
• 18:36
  independent variable is study time, meaning that it's
• 18:40
  independent of anything else.
• 18:44
  And my dependent variable is exam scores,
• 18:48
  because what we think is going on is the dependent variable depends
• 18:53
  upon changes in the independent
• 18:57
  variable. Okay, so it doesn't— it can't change on its
• 19:00
  own. It has to have something as you would say,
• 19:03
  pushing on it or pressing on it. Okay. So how
• 19:09
  do we determine this? Well, maybe what we do is we
• 19:14
  have some students study for
• 19:18
  1 hour a week, 4 hours a week, and let's
• 19:22
  go 8 hours a week. So we have 3 groups. Okay.
• 19:29
  And then, so we have them do this study time and we have these 3
• 19:34
  groups, and then we have them do this exam.
• 19:37
  Okay, now if our hypothesis is true,
• 19:42
  what we should see is that the people who
• 19:46
  studied 8 hours did significantly
• 19:50
  better on the exams than the persons
• 19:54
  who did the 4-hour and the 1-hour,
• 19:57
  and the person in the 4-hour should score better than the person who studied
• 20:01
  1 hour if, if, if our hypothesis
• 20:04
  is true. Okay, now to
• 20:08
  determine this type of test, we usually look at either a t-test or
• 20:13
  an ANOVA. Okay, a t-test
• 20:18
  is used when we only have 2 independent variables.
• 20:21
  So if we just did studies for 1 hour,
• 20:24
  studies for 4 hours, we would do a t-test, and that test
• 20:28
  would compare to see whether those groups are
• 20:32
  significantly different from each other.
• 20:37
  If we have more than 2 groups, we do an ANOVA,
• 20:41
  which stands for analysis of variance.
• 20:45
  So in this case, we have 3 groups. And so what
• 20:49
  we will do is we'll run what's called an ANOVA test. And again, we'll go
• 20:53
  through these in very much detail to see
• 20:57
  whether or not these groups actually do differ from each
• 21:01
  other. And if they do, and let's say
• 21:05
  the 8 does come out that, you know, people who studied for 8 hours
• 21:09
  did really much— did much better than the other 2
• 21:12
  groups, we can then at that point say study
• 21:16
  time does influence exam
• 21:20
  scores. Okay? So that's the— this is very
• 21:23
  basic. This is very basic. And we'll
• 21:27
  have more details in it later. Okay, so just to
• 21:32
  kind of mention, I mentioned qualitative methods as a potential
• 21:36
  ground. And some of you, if you go into business
• 21:40
  consulting and stuff like that, you'll definitely be doing this type of analyses.
• 21:47
  But let's really quick review the
• 21:51
  difference between quantitative methods and qualitative methods.
• 21:55
  Quantitative methods are when I want to know the objective
• 22:02
  reality that's going on. It's, it's when I want to utilize
• 22:08
  numbers to describe a phenomenon.
• 22:11
  Okay, and it's always usually group-based, meaning that
• 22:16
  quantitative methods do not do well if you only have a
• 22:20
  sample size of 4 or 5 people. In fact, the standard
• 22:24
  for quantitative methods is you should at least have 30 people or more.
• 22:30
  And you're really just trying to get at the objective reality that's going
• 22:33
  on. Okay, qualitative methods are
• 22:37
  used when you want to know the subjective reality of
• 22:41
  something. Or someone or a group, meaning what
• 22:45
  you want to know is what is their lived experience
• 22:50
  in a given situation. Okay.
• 22:53
  Mixed methods is when we combine these 2.
• 22:57
  All right. So let me give some examples
• 23:01
  of mixed methods using qualitative
• 23:04
  methods. Okay.
• 23:07
  Mixed methods I use them when I
• 23:11
  used to do organizational consulting.
• 23:15
  So the main thing that I was hired for was mainly
• 23:18
  nonprofits who were struggling to meet some grant goals,
• 23:22
  or they were having organizational problems and the like.
• 23:26
  So my job was to go in and analyze the situation
• 23:29
  and then provide recommendations on how the organization should move forward.
• 23:36
  My usual approach, because a lot of it usually deals
• 23:39
  with the human factors in a situation and either
• 23:44
  human error or just human neglect.
• 23:47
  So the first thing that I would do is I would always
• 23:51
  start with a survey. And I would survey about
• 23:56
  customer service, for example, job satisfaction,
• 23:59
  organizational commitment.
• 24:03
  Motivation, all of these different types of quantitative methods.
• 24:08
  Okay, and I would measure all of those things and I would put them into
• 24:11
  the statistical software and it would shoot out at me that, hey,
• 24:15
  these are the issues that are going on.
• 24:19
  Okay. And so from that point,
• 24:23
  I'm drawing a picture of the organization at the organization level.
• 24:27
  How much burnout is there? How much job satisfaction is there?
• 24:30
  What is the the work productivity.
• 24:35
  Okay, so now I have an idea. So let's say my
• 24:38
  objective methods led me to find
• 24:43
  out that they have one, let's say, a really high turnover rate,
• 24:47
  and their burnout rate seems to be really high,
• 24:50
  especially with newer type of employees.
• 24:54
  Okay, so I have those objective information
• 24:58
  about the organization on the organizational level.
• 25:01
  And those are the— that those are kind of some of the questions that you'll
• 25:05
  be using these tests that you'll learn about in this class to figure
• 25:09
  out. Okay. But the next step for me is,
• 25:13
  okay, I have these, these things
• 25:18
  on the organizational level going on, high burnout,
• 25:22
  low motivation, blah, blah, blah. But what I don't know is
• 25:27
  I don't know what those employees' lived reality in
• 25:31
  that environment with those, those issues going
• 25:34
  on. So the next step in organizational consultation for
• 25:38
  me was to then go do interviews with the
• 25:42
  employees and try to determine what is
• 25:45
  going on and, and really do
• 25:49
  good interviews. Okay,
• 25:53
  usually I would do some observations. I'd go sit in someone's office
• 25:57
  for a while and just watch And I would
• 26:01
  always be using what's called the grounded theory approach. And let me explain
• 26:04
  these 3 things since we're doing this on the group level.
• 26:08
  Grounded theory is when you're using a, a structure,
• 26:14
  a platform to work on in your analyses.
• 26:18
  So in my situation, I'm using grounded theory
• 26:21
  because I'm going into these interviews and these observations
• 26:25
  with this knowledge that this organization has high burnout,
• 26:28
  high turnover, among newer employees. So that's
• 26:32
  my grounded theory, that there is something going on with this
• 26:36
  group in this organization. So then I'll
• 26:40
  use this theory to determine, okay, who do I need to interview?
• 26:45
  Who do I need to go do observations with? Okay.
• 26:49
  Now, this seems like a lot of work for an organization,
• 26:53
  right, doing all this analysis and stuff like this,
• 26:56
  but In— when you do this kind of quality type
• 27:00
  of analysis, starting with the objective and then going and trying
• 27:04
  to figure out what the lived experience is and doing it methodically,
• 27:08
  you actually end up saving organizations a lot of money.
• 27:11
  In this situation, this was actually a real situation that I dealt with,
• 27:15
  with a mid-sized hospital. And after
• 27:18
  all of that analysis, what I found out is, is that there were 3
• 27:22
  individuals in that organization who were triangulating all
• 27:26
  the new employees. And what they would do when a new employee
• 27:30
  came in, these 3 individuals would decide if they liked that person or not.
• 27:34
  And then they would go do destructive behaviors if they didn't like the
• 27:38
  person. They would start doing very destructive behaviors
• 27:42
  with the individual and they would triangulate each other. And it
• 27:46
  just, that's what created that whole situation.
• 27:49
  So originally this hospital was planning on spending $50,000
• 27:53
  to $100,000 on on training materials,
• 27:57
  on motivational materials. And really all they had to do
• 28:01
  was fire 3 people. The organization is doing wonderful.
• 28:04
  Well, last time I've talked with them, they were doing wonderful. That was about 5
• 28:07
  years ago. But that's what I mean by doing a mixed method.
• 28:12
  All right. And so, yeah,
• 28:15
  so that's kind of how that works. I'm going to stop right here and ask,
• 28:19
  does anybody have any questions at this moment? I know I'm going over
• 28:23
  this material really quickly. But again, this is just
• 28:26
  a starting point.
• 28:32
  I gotta turn off my pen real quick.
• 28:39
  Alright, so is there any questions at this moment?
• 28:56
  OK, great. Thanks, Christy, Caleb,
• 29:00
  Diana. All right, thank you.
• 29:04
  OK. All right, so let's move on and continue talking more about
• 29:08
  quantitative methods, all right, and get
• 29:12
  more into this more deeply.
• 29:15
  Thank you, Brittany. All right, so again,
• 29:19
  quantitative method is when we have the intent to measure something to answer
• 29:23
  some type of question. Okay, in the scientific field, we call this a research
• 29:27
  question. And usually we just ask,
• 29:31
  what is it that I want to know and why is it— why is the
• 29:35
  question significant? Okay, and so once we have the
• 29:39
  question, we develop a method. Okay,
• 29:42
  and these are some of the things that we need to spend some time talking
• 29:47
  about is, is, you know, sample,
• 29:50
  population, procedures. All of those will determine what
• 29:54
  type of statistical analysis that you'll be doing.