Saturday, October 25, 2014
Grades have been posted. While I no longer
hold office hours, I am on campus on and off throughout the summer. You're
always welcome to stop by my office to speak to me about the course or grad
school or other future academic plans (I just don't have a set schedule). I'm
also available more regularly in the fall.
It was a great semester, enjoy your summer,
and I hope to see you in future courses.
Chris:-)
Final exam review
Here are some review questions to help you study. The questions have multiple parts and I’ve
indicated point values to help organize a response. Remember I have extra office hours on Monday and Wednesday.
Good luck studying!
===========
Ch 6—Levels of
Measurement
You begin with the concept of “education” and must now
operationalize it. Create questions that
measure education at nominal, ordinal, and ratio levels. Provide response categories. (6 points) (Hint:
Here’s how to approach this question.
You’re asked to write both questions and responses for each level of
measurement. There are three levels of
measurement.)
Ch 7—Sampling
I am interested in the voting behavior of young adults (ages
18-24) in Canada. I have limited
resources, so I’m going to confine my study to a survey of York undergrads in
this age group. However, this still
amounts to over 40,000 people! I can’t
survey that many, so I’ve decided to randomly sample 400 students. What are the population and target
population? What are the sampling ratio and sampling
elements? What is the sampling frame and how might I obtain it? (6
points)
Ch 8—Survey Research
In tutorial we saw a news article. It reported that “68% of Torontonians were
opposed to forcibly removing the homeless from public places.” Presumably, what was the question asked? What’s wrong with how it was phrased? How would you fix it? (3 points)
Ch 9—Experiments
Ontario has the highest rate of workplace injury. A group of researchers has developed a
training module specifically designed to teach people how to avoid repetitive
strain from activities like typing and how to lift heavy items without
straining the back. The module must be
tested via experimental design before being adopted by the province. Two workplace offices have volunteered to
participate in a study. Describe the key
elements of your experiment and explain how you can conduct a double-blind
study (Hint: The elements include variables, pre and post-tests, etc.) (8 points)
Ch 10—Content
Analysis
Think back to the in-class content analysis of film
sequences from Requiem for a Dream
and Trainspotting. What was the unit of observation? What was the unit of analysis? Explain how the analysis was “structured”
(frequency, space, intensity, or direction).
Ch 10—Content Analysis
Imagine you and I are asked to work on a content analysis of
newspaper comics. The research question
is “How is violence portrayed in weekly comic strips?”
1. What is the unit of
analysis? What is the unit of observation? (2 points)
2. Provide an example of a manifest code. Explain
exactly what has to be seen for the comic image to qualify as manifest content.
(2 points)
3. Provide an example of a latent code. Explain exactly
what has to be seen for the comic image to qualify as latent content. (2
points)
4. You and I rate the manifest and latent content of a bunch
of comics. Our percent agreement is 90%.
That’s really high! Or, is
it? Are there any potential problems
with the apparent success of our ratings? (2 points)
Key Points: Chapter 15 Analyzing Qualitative Data
Analyzing qualitative data—i.e, coding—has many of you
frustrated. It shouldn’t because it’s
really no different than what you’ve already in essays for other
courses—namely, present a reasonable argument based on evidence.
By way of quick overview, we have three basic stages of
coding: open, axial, and selective. Keep
in mind my “funnel” metaphor from lecture:
each stage of coding is meant to distill or refine the bulk data into a
more concise understanding of the phenomenon under study.
Open codes are the
first pass through qualitative data where we “boil down” or summarize the
information. Remember we will have LOTS
of open codes. And remember that we open
code with reference to your specific research
question.
Axial codes are
where we begin to make linkages between open codes, either by grouping together
like codes or by organizing codes that are related in some pattern. For example, in the lecture slides we noticed
that “loneliness” and “desire for company” were like codes, while “conflict”
and “moving out of parents house” were patterned because one led to the other.
Selective coding
is where we try to identify a core category around which the bulk of the
information will fit. The goal is to
locate a concept or theme that explains a good chunk of what’s going on in the
axial codes. There could be more than
one selective code since transcripts cover a lot of material.
Once we’re done coding, we turn to analytic memo writing,
which basically means we develop a theoretical explanation. Now we’re stepping back to reflect upon and
interpret what we see in our data thanks to these successive coding
stages. Ideally, we produce some kind of
general causal explanation, however rough it may be. The
important point is that the explanation is supported by our coding; it must
also make sense in the context of the larger interview data or field notes.
It’s the last part that has many of you especially
concerned. This is why I started this
entry by saying that this coding process is much like any essay you’ve ever had
to write: make a logical argument based
on evidence that fits the topic. You’ll
be evaluated mostly on your understanding of the process and how well you
demonstrate systematically coding qualitative data. How well you’re able to reflect upon that
process and present an analysis is a much smaller part of the evaluation.
Five things to know about Pearson r
The final topic for Chapter 11 is the Pearson r coefficient. You won’t be asked to
calculate this statistic, but you will need basic conceptual understanding of
it. You’ll need to supplement this blog entry with the textbook because I can’t
include diagrams here.
1. The general idea
The Pearson coefficient (often referred to as “r”) is a measure of bivariate correlation. This means it measures the
strength of a relationship between two variables. It does NOT measure causation
(remember there are three criteria for causation and correlation is only one).
For instance, it seems intuitive enough that there is a
positive relationship between the variables annual income and years of education
(the more money you make, the more education you likely to have). Therefore
we’d expect a Pearson coefficient to indicate a strong relationship between
these two variables. By contrast, it’s hard to imagine that there’s a
relationship between the variables eye
color and income. It doesn’t make
sense that the two have anything to do with one another. In this case, we’d
expect our Pearson coefficient to indicate either a weak or nonexistent
relationship.
Now let’s talk about specifics.
2. The coefficient
The Pearson coefficient ranges from -1 to +1. The closer the
value is to -1 or to +1, the stronger is the relationship between variables.
Negative and positive values that are close to “0” indicate a weak relationship
between variables. You’ll recall that we have two kinds of relationships
between variables, negative and positive. Those relationships are
reflected in the Pearson coefficient, which is why both -1 and +1 indicate a
“strong” relationship.
3. Type of variable
Pearson r can only
be used to measure variables at the ratio
level of measurement. Nominal and ordinal variables are null and void. The
short and sweet of it is that the Pearson coefficient relies on a calculation
of the mean. And as you already know, the mean can only be calculated for ratio
level variables. So, a red flag should go up if you’re asked to interpret a
Pearson coefficient for the variables age and gender. This would be an invalid
use of Pearson r because gender is a
nominal variable.
4. Type of
Relationship
Pearson r can only
be used to measure linear relationships.
Curvilinear relationships make the Pearson statistic null and void (the
curvilinear relationship between the variable may be real, it’s just that the
Pearson statistic cannot be used to measure or evaluate it). In lecture, I used
“income” as an example of a curvilinear relationship. I said that income over a
lifetime is not a straight line; most people make no money as a child, lots of
money in their prime, and then minimal income after retirement. If you can
imagine plotting that relationship on an x/y graph, you’d have a curve. Another
curvilinear relationship is between health and age; the health of children and
the elderly tends to be poorer than the health of young and middle-aged adults.
Again, you’d have a curved plot on a graph. You may be asking: How do I tell if
a variable is linear or curvilinear? The short answer is that you’d actually
have to plot it and look for a visible pattern. But don’t worry about that, for
our purposes you simply need to know that Pearson r is only appropriate for linear relationships.
5. Interpreting
examples
Interpreting a Pearson coefficient is simple as pie. A
relationship between variables can be: a) weak, b) moderate, or c) strong.
You’ll have to double-check this in the book (I don’t have mine handy) but the
guideline is something like 0-.3 is weak, .31-.69 is moderate, and .7-1 is
strong (same for negative numbers). So if age
and years of education have a Pearson coefficient of .4, you’d conclude that a
moderate positive relationship exists (the older you are, the more education
you have). If amount of smoking and life expectancy in years have a Pearson
coefficient of -.8, you’d conclude a strong negative relationship exists (the
more you smoke, the fewer years you live). If income and IQ have a
Pearson coefficient of “.1”, you’d say a weak relationship exists (so weak, in
fact, you’d probably conclude that no relationship exists).
That’s it. Those are the five key points to know about
Pearson r. To close, here’s a quiz to
test your understanding. Bring questions on Monday.
- Interpret
a Pearson’s coefficient of .75 for the variables age and number of children.
- How
would you draw a Pearson’s coefficient of “0” on a scatter plot?
- True
or False: A coefficient of .5 is a stronger indicator that your hypothesis
is correct than a -.5 coefficient.
- True
or False: A coefficient of -.75 for age and religion indicates a strong
negative relationship between age
and religion?
- True
or False: For the ratio level variables age and income, a
coefficient of .8 means there is a strong causal relationship between age and income.
More on Standard Deviation
The idea of SD is that you want to know how much variation
there is in your data around the mean. The smaller the SD, the more closely the
data are concentrated in the distribution; the larger the SD, the more widely
or loosely distributed are the data in the distribution. In lecture, I referred
to the SD as the “average freak.” Freaks are abnormal and by definition are not
average and do not conform to the mean. So the standard deviation tells us how
far from the mean value the “average freak” is.
Here’s an example. One distribution has a SD of 6 and a
different distribution has a SD of 20. A SD of 6 is much smaller than an SD of
20. That means that the data in the first distribution are much closely grouped
than the data in the second distribution.
Remember that the mean is simply the average score. For
example, the average of 20 and 0 is 10 (twenty plus zero divided by two equals
ten). The average of 11 and 9 is also 10 (eleven plus nine equals twenty
divided by two equals 10). But in the
second case our data points range from 9 to 11 and are much more closely
concentrated around the average (10) then in the first case where our range
from 0 to 20. Therefore, the SD in the second case would be a much smaller
number because the values are more closely concentrated around the mean.
In-class example:
Today we looked at test scores. I told you that both of my
classes have the same average of 70%, but that they have different SDs.
Tutorial 1 had a SD of 15% and Tutorial 2 had a SD of 10%. This means that
Tutorial 1 scores ranged between 55-85% (or 15% plus and minus 70%). Tutorial 2
scores only ranged between 60-80%. From this we would conclude that test scores
in Tutorial 2 are more closely grouped around the mean; therefore, the mean
score of 70% is actually more representative of students in Tutorial 2.
Let me help you study...
1. A
research decides she wants to study new immigrants to Canada and their
experiences in the Canadian labour force.
From this general topic, there are two ‘concepts’ she has to
defined. What are they?
2. She decides to do a quantitative study of
the above topic. After she has decided on her concepts, WHAT does she need to
do to measure them?
3. The researcher is going to do a survey on
the above topic. She needs to construct good survey items. What are some items
she could ask people to determine their immigrant status? Include all
categories of the variable.
4. The researcher made a variable for immigrant
status and found that other researchers thought that it looked like a pretty
good way of measuring the concept. What kind of validity does that demonstrate?
5. She also found her variable was pretty close
to how another famous researcher was measuring immigrant status. What kind of
validity does that demonstrate?
6. The measure also seems to fit pretty close
with its conceptual definition. What kind of validity does that demonstrate?
7. The researcher walks around York university
and asks random people her survey questions.
What kind of sampling is that?
What is a disadvantage of that kind of sampling?
8. If the researcher did a simple random sample
of York university students, she would need a list of all elements of her
population, which is known as a _______________. Where could she get such a list?
9. After getting some surveys completed, the
researcher has some data to analyze. She wants to get some measures of central
tendency. What measures of central tendency can she obtain for a variable that
assesses whether someone is an immigrant or not?
10. She also asked respondents how many hours
per week they worked and obtained data on the number of hours they spent in
paid employment. What measures of central tendency could she examine there?
11. She found that immigrants had an average of
30 hours per week in paid employment, while non-immigrants has 32. The standard
deviation for immigrants was 15, but for non-immigrants it was 2. What does
this mean roughly?
Test your understanding: Measures of Central Tendency and Variation
Understanding measures of central tendency and variation (or
dispersion) are key to analyzing quantitative data because they enable us to
summarize large quantities of data.
Imagine you are asked to describe this data: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9
What can you say about this data? How it be summarized? By the end of this post, you’ll be able to
answer these questions in several ways.
Measures of Central
Tendency: measures of tendency tell us what our values tend to be on
average. There are three: 1) mode, 2) median, and 3) mean.
The mode is the
most frequently occurring case. It is
the most general and least precise measure of tendency; it can be used with any
level of measurement. A sample of smokers
reveals that 10 smoke Camels, 15 smoke du Maurier, and smoke Dunhill. The mode is Dunhill (not 20) because it
occurs most frequently. If 10 students
scored 60 on the midterm, 15 scored 70, and 5 scored 80, then the mode is 70.
The median is the
midpoint in a data set. Once you have
rank ordered the cases, the middle value is the median. It can only be determined for ordinal and
interval/ratio levels (nominal data cannot be rank ordered). For example, 5 is the median of these data:
3,3,4,4,5,6,6,7,7 (if there were an even number of values, then the median
would be the average the middle two).
The mean is the
average value. It is determined by
adding up the individual cases and dividing by the total. 5 is also the mean for the above data. It can only be calculated for interval/ratio
data (nominal and ordinal data lack equal distance between categories).
The advantage of the mean is that it’s technically the most
accurate estimation of data because it takes into account every value, unlike
the median that only represents a single value.
The disadvantage is that the mean is sensitive to outliers (extreme high
or low values) and can be misleading, whereas the median is unaffected by
outliers. As a rule of thumb, it is
better to rely on the median if you know or suspect outliers in the data.
Measures of tendency are easy enough to figure out, but
they’re of limited use because most values vary to some degree from the mode,
median, and mean. For example, the mean
of 9 and 11 is 10, but so is the mean of 0 and 20. Here’s a better visual example:
Data set 1: 1, 1, 2,
5, 8, 9, 9
Data set 2: 2, 4, 4,
5, 6, 6, 7
5 is the mean value of both data sets. But as you can see, there’s a lot more
variation in the first set than in the second and so the mean actually better
reflects the second data set. These data
sets are small and easy to grasp visually, but imagine if you were dealing with
thousands of values! Then it wouldn’t be
so easy. Fortunately we have measures of
variation to tell us how spread out are the data.
Measures of Variation:
There are several measures of variation but we’re focusing
on three: the range, interquartile range (IQR), and the standard deviation
(SD).
The range is easy
enough—it’s the span of values listed lowest to highest. The range for the first data set is 1-9 and
the second is 2-7. It’s a vague measure,
but nonetheless gives us some sense that the values in the first data set are
more spread out than in the second.
The interquartile
range (IQR) tells us how closely the data are dispersed around the
median. There’s a formula for
calculating IQR, but the short and sweet of it is this: First, order the data least to highest, next
identify the median, then divide the data into four sections (or quartiles),
and finally drop the first and four quartiles (the highest and lowest
values). What remains are the second and
third quartiles separated by the median.
The IQR tells us how closely dispersed half our data are around the
median.
Consider data set 3:
1, 2, 3, 4, 5, 6, 7, 8
The data are ordered, the median is 4.5. We drop the first quartile (1, 2) and the
fourth quartile (7, 8), and we’re left with an IQR of 3-6. Now we know our median value is 4.5 and that
the middle half of the data are between 3-6.
The standard deviation
(SD) tells us how closely the data are dispersed around the mean. It is the most accurate measure of
variation. There’s a formula for
calculating SD that is relatively straightforward but involves several
steps. I’m only going to talk conceptually
about SD, so you’ll need to refer to the text for learning how to calculate it.
As with the mean, the benefit of SD is that it incorporates
all values into the calculation and therefore is more representative of the
entire data set; the disadvantage is that, like the mean, SD is also sensitive
to outliers.
Let’s say that our mean value is 20 and that we know a
single value in our data, 17. How close
is this value to the mean compared to the rest of the values in the data
set? 17 seems close to 20, but if most
of our data points are 18 and 19, then 17 is actually far from the mean
compared the majority. Fortunately, SD
comes to our rescue!
If SD=5, then we know that on average the values in our data
set are within 5 points of the mean. In
this case, we would conclude that a value of 17 is close to the mean of
20. By contrast, if SD=1 then we know
that on average values are within 1 point of the mean. In this case, we’d conclude that 17 is far
from the mean and therefore not representative of most values.
On a final note, SD is related to the idea of the “normal”
or bell curve, where the mean is at the center of the distribution and the SD
represents sections of data to either side.
Data are usually located within three SDs above and below the mean. For example, if SD=5 with a mean of 20, then
we know that the first SD is between 15-25, the second SD is between 10-30, and
the third is between 5-35. In this case,
a value of 17 would be within 1 SD of the mean whereas a value of 7 would only
be within the third SD. And “outlier” is
any value beyond three SDs of the mean.
Chapter 11 Summary
As you know, Wednesday’s lecture was interrupted by a fire
drill. Ugh, how inconvenient. We must, however, move forward so below I’ve
provide a conceptual synopsis of Chapter 11.
============
Up to now we’ve looked at methods of collecting quantitative
(numeric) data. Surveys, experiments,
and content analysis are all methods that generate this data. And we looked at different ways to create
measures for this type of data, namely nominal, ordinal, and ratio levels.
Now we’ve come to Chapter 11. This chapter is about making sense of the
numeric data that we collect. It teaches
us how to describe data generally instead of on an individual basis. We do this all the time in our lives. For example, suppose you’re asked to describe
how good of a student you are. You
wouldn’t list off every single grade you’ve ever received in school. That would be ridiculous. Instead, you’d give some kind of summary
statement, like describing yourself as an “A” or “B” student, or by giving your
overall GPA. This idea of descriptive
summary of data is the focus of Chapter 11.
First, think of data
as a group of information about something.
If I wanted to estimate the age of 2030 students, then I’d ask a bunch
of people, not just one. And that bunch
of people that I asked would provide me with a pool of information to work
with, or a “dataset.” Once we have data,
then we have to make sense of it.
Second, groups of data can be analyzed individually or they
can be compared to other groups. Univariate analysis examines a single
group of data that are represented as one variable. For instance, age is a variable that we can
analyze all by itself (e.g., the average age of students in Soci 2030). Or, we can use bivariate analysis to compare age with another variable, like
gender (e.g., the average age of women in 2030 compared to the average of
men). In this post, I’m going to deal
only with univariate analysis.
Third, we talk about data two ways: 1) measures of central tendency and 2)
measures of variation (or dispersion). Central tendency is simply what data
points “tend” to be, or their average.
For example, the average age of students in 2030 is 21. However, that’s just a tendency; in fact,
many students are not 21. Measures of dispersion tell us how
closely most of the cases in our data are to the average. For example, most students in 2030 are within
1 year plus or minus of 21. This gives
us two related pieces of information: we know that on average 2030 students are
21 years old and we know that most of those who are not 21 are still between
20-22. In this example, the average and
the dispersion are pretty close. But
this is not always the case.
Here’s a different example.
Consider the age difference between students registered at York and students
at Atkinson. York is set up on the
traditional university model and therefore students tend to be quite
young. York students are usually fresh
out of high school, live at home or in residence, receive partial or full
financial support from parents, and have limited professional work
experience. Not surprisingly, York
students tend to be about 20 years old on average, give or take two years. There’s not a lot of variation in age.
Now contrast York students with Atkinson students. Atkinson was originally set up to accommodate
mature students. Mature students are
older on average for a variety of reasons.
Maybe someone worked for a few years after high school and then decided
to go to university. Or, maybe they
worked for 20 years, realized that they couldn’t get promoted further without
more education, and are now attending night school to earn extra credits. Or, maybe they worked their whole lives,
retired, and decided to go to school and earn that degree they always dreamed
of. In short, Atkinson students tend to
be older and are much more diverse in age.
The average age is 35, with most between 25-45 years. Compared to York students, Atkinson students
have a higher average age, plus there is also more variation from that average
(plus or minus 10 years).
In short, central tendency gives us a general idea about
what our data tend to be, while measures of dispersion help us understand how
much data vary from that tendency.
Fourth, now that we have the basics, we need to identify the
different measures of central tendency and variation. Measures of central tendency are the mode,
median, and mean. Measures of variation
are the range, percentiles, and standard deviation. The level of measurement of a variable
(nominal, ordinal, ratio) determines which measure of central tendency or
variation can be used.
In tutorial we collected data on the kind of drugs people
used over the weekend. This was a
nominal level of measurement and therefore we could only identify a mode, or
most commonly occurring value. Next, we looked at data on 15 quiz scores. This was a ratio level measurement and
therefore we were able to identify a mode, median and mean.
Measures of variation require an ordinal or ratio level
measure because the data must be ranked before we can describe how it varies
(remember nominal data cannot be ranked).
We’re only focusing on the range and standard deviation. The range
is a listing of data from smallest to largest value. For example, the range in age of 2030
students is 18-28 years old. This
doesn’t tell us much. We know the age of
the youngest and oldest students but we don’t know the variation of ages. For example, there could be nineteen 18-year
olds and one 28-year old or there could be 2 students who are 18, 19,
20…28. Either way the range would be the
same.
The standard deviation
(SD) is more precise and tells us how far from the mean that our data
vary. Because the mean can only be
calculated using ratio level data, this implies that SD also requires ratio
data. Remember that you will not need to
calculate SD. It is sufficient to simply
know that it tells us how varied are our data—the larger the SD, the more
variation in the data. One SD refers to
about 2/3 or 66% of our data, two SDs refers to about 95%, and three SDs refers
to about 99% of the data. For example,
the mean annual income in a dataset is $50,000/year with an SD of $5,000. One SD is plus or minus $5,000 of the average
value. This means that approximately 2/3
(or 66%) of people earn between $45-$55,000/year. Two SDs means that approximately 95% earn
between $40-$60K/year, and so forth.
Stay tuned for more!
Key Points: Chapters 9 & 10
First, some house
keeping.
1) Monday is review day. The format will be the same as
before where you ask questions of one another. I distributed a review sheet,
but also bring your own questions (b/c what’s on the review sheet won’t be on
the test!).
2) I will hold office hours after Monday’s lecture. If lots
of people show then be prepared to share me or limit your questions. If nobody
else shows, then I’m all yours.
Now a recap. The recipe for this week was experiments and
content analysis with a dab of validity and a pinch of official stats.
Delicious!
Regarding experiments. Knowing the components of
experimental design is only the beginning. You also need to understanding how
experimental research can be compromised by different threats to internal validity. I also sent additional practice
questions for internal validity through Moodle (some are tricky, so pay
attention). The difference between laboratory
and natural experiments was also
emphasized in lecture and tutorial. Ask yourself: How do internal and external
validity compare between natural and field experiments? Why does this matter to
methodologists? In lecture, I also touched on statistical regression (remember the “smart and dumb babies”
example?).
Regarding secondary analysis. Here we focused on content
analysis, a non-reactive method for
systematically analyzing texts. There are different techniques for coding text (frequency, space, intensity) as well as
types of codes (manifest, latent).
You had two exercises: one on drug use in film and the other on gender in
magazines. There’s no point memorizing specific answers to those exercises
because the test will have different examples. Instead, study by applying the
process to new examples. For instance, what if you had to conduct a content
analysis of “anger” in film? Would you “count” anger using frequency, space, or
intensity? What are some examples of manifest vs. latent representations of
anger?
Finally, in lecture I talked about the use of existing statistics in secondary
analysis. I emphasized the importance of social
context in accurately interpreting rates and trends. I distributed a
similar review exercise on youth homelessness. It asks you to interpret rates
and trends and to speculate about social context. Rates refer to instances at a given moment (e.g., rates of
homelessness in 1990) while trends
refer to changes over time (e.g., trends in homelessness for different groups
over many years). It might help to think of them like cross-sectional and
longitudinal time dimensions.
Good luck studying. See you Monday.
Test your understanding: Survey Questions
Creating survey questions is trickier than you might think. Here are more examples
on survey design. You can also google "survey questions" or "bad
surveys" or something along those lines and examples abound! Take some time
to work through the examples—practice makes perfect!
http://www.statcan.gc.ca/edu/power-pouvoir/ch2/exer/5214910-eng.htm
http://www.statcan.gc.ca/edu/power-pouvoir/ch2/questionnaires/5214775-eng.htm#a15
Test your understanding: Measurement and Sampling
As promised, here are
the quiz questions you worked on in groups. For those absent on Monday, the
questions are drawn from Ch 6 & 7 material, and are similar to the kind of
test questions that I typically ask (a stronger hint, I cannot think of). I
hope they are helpful when studying. Come see me if you still have
questions—good luck!
1. Two sociological
constructs are income and higher education. Working from a quantitative
perspective (as we’ve been doing in class), I want you to first conceptualize
the constructs and then operationalize them (4 pts). Specify which
construct will be your independent variable and which will be the dependent
(1 pt). Now present a conceptual hypothesis linking your constructs (1
pt). Finally, develop a testable empirical hypothesis for the variables
(2 pts). See pp. 108-10
2. I want to compare
levels of homophobia between people who live in downtown Toronto and those who
live in GTA suburbs. I’ve talked with my peers about this research project.
They agree with me that homophobia is an irrational fear of gay and lesbian
people. The existing literature is even more detailed in specifying that
homophobia is an intense disliking of gays and lesbians as opposed to “fear”
per se; the literature also discusses acts of overt discrimination toward gays
and lesbians. I’ve decided to include all three as indicators in my measure of
homophobia. This will be my dependent variable. One existing study successfully
used “church attendance” during childhood as a predictor of homophobia; it
found a correlation between high church attendance in youth and increased
likelihood of eventually disliking gays and lesbians. I’ve decided to use this
same measure as one of my independent variables. I’ve also created a related
measure, “religiosity,” or how religious people are; I believe it is similar
because people who are more religious are likely attend church more often. I
predict both will correlate with the dependent variable. I decide to pilot test
my measures in a survey of two local political groups. Interestingly, the pilot
results showed that political conservatives showed higher levels of homophobia
than the political liberals. Which of the four types of measurement validity
have been met and which have not (4 pts each)? Explain why (4 pts each). See
pp. 114-5
3. I am interested in
the voting behavior of adults aged 18-22 in Canada. Because I have limited
resources, I’m going to confine my study to York undergrads. However, this
still amounts to 10,000 people! I can’t survey that many, so I’ve decided to
draw a sample (n = 100). What are the population and target
population? What are the sampling ratio and sampling elements?
What is the sampling frame and how might I conceivably obtain it? (6
pts) See pp. 141-3
4. Similar to the
scenario as above. I am interested in the voting behavior of young adults aged
18-22 in Canada. I have two research questions: 1) what percentage of young
adults voted in the last federal election and for whom did they vote, and 2)
how did deciding to vote or not to vote emotionally affect them? The first
research question is quantitative in nature while the second is qualitative.
Due to limited resources, I will confine my study to York undergrads. I want to
conduct one probability and one nonprobability sample. For each research
question above, explain why it would be better to use a probability or
nonprobability sample (2 pts). Now choose an appropriate type of probability
sample and describe how it would be applied to one of the two research questions
(3 pts). Do the same with a nonprobability sample for the other research
question (3 pts). You have many options but NO convenience samples! See pp.
137-55
Test your understanding: Margin of Error
Today in lecture I touched on margin of error. I want to
make sure this was clear because it will be on your next exam. Below are two
examples of interpreting margin of error.
===============
--Example 1--
According to a recent Globe and Mail poll, 70% of
Torontonians would vote Liberal if a federal election were help today. The poll surveyed 500 people living in
Toronto on February 3 with a margin of error of 5 percentage points. It’s considered accurate 19 times out of 20.
1. Interpret the
margin of error.
If given again, the same poll would have results within plus
or minus 5 points of 70 percent (or between 65-75%) 19 times out of 20. Only 1/20 polls would have results outside
the 65-75% margin of error.
===============
--Example 2--
OTTAWA—The Conservatives, hit by bad publicity in the wake
of Stephen Harper’s second effort to suspend Parliament, have slipped against
the Liberals in a new poll released Monday but still maintain a 6 percentage
point lead.
The CBC poll taken February 8 of 1095 registered voters
showed the Conservatives would maintain their minority government if an
election were held now. Public support
for the Conservatives is 43%, with the Liberals 37%, NDP 9%, the Bloc trailing
at 7%, and 4% of voters undecided. The
poll has a margin of error of plus or minus four percentage points 95/100.
A similar poll on January 11 showed the Conservatives with
46% support, Liberals 34%, NDP 14% and Bloc 6%.
1. What is the margin of error?
Plus or minus 4 percentage points 95/100 times. For every 100 polls of the same question,
results for 95 will be within 4 points.
Put another way, only 5/100 (or 1/20) polls will be completely
off-the-wall different!
2. Is there reasonable statistical basis for
claiming that favorability for the Conservatives has declined?
No, statistically there is no change. If you gave this poll 100 times, 95 of those
times the percentage of people giving a particular answer would be within 4
percentage points of this poll. The percentage
of people in the latest poll who say they support the Conservatives is within 4
percentage points of those who said they supported them in the previous poll
(43-46%) and the same goes for the Liberals.
Therefore the Liberals have made up no measurable ground on the
Conservatives.
3. What conclusions can be drawn?
There results are more or less consistent. If anything, support for the NDP has
measurably dropped.
Key Points: Chapters 7 & 8
As we enter into reading week, I want to
provide an overview of what we’ve discussed so far. In particular, I want to
make sure it’s clear how this material builds on what we learned about
measurement in Chapter 6.
Chapter 7 is about probability and non-probability
sampling. Remember that probability sampling techniques are associated with
quantitative research, while non-prob techniques are associated with qualitative.
In lecture, I also discussed margin of
error, which has to do with how accurate we believe our probability sample
to be. I distributed a handout with sampling exercises on one side and margin
of error exercises on the other. Finally, we discussed recent research on oral
sex as an example of the “science of the
sophomore” (those who were absent really missed out!).
Chapter 8 is about surveys. We looked at a
poll on attitudes toward the homeless and identified problems with question
wording. Lots more examples can be found online if you want additional
practice. While we focused on identifying and fixing “bad questions,” there is
additional Ch 8 material that you’re responsible for knowing (e.g., types of
surveys, pros and cons).
We’ll talk about experiments when we
reconvene. For now, enjoy reading week.
More on measurement!
In lecture, I stressed that
some variables can be measured with more precision than others—hence there are
“levels” of measurement. And the
level(s) at which we can measure depends on the nature of the variable—i.e.,
how it has been operationalized (this
is important later!)
There are two types of
variables (discrete & continuous) and four levels of measurement (nominal,
ordinal, interval & ratio). Discrete variables are fixed and
descriptive while continuous
variables have infinite attributes. Each
type of variable contains two levels of measurement. Discrete variables are nominal or ordinal while
continuous variables are interval or ratio.
Nominal
measures contain attributes or “qualities” that are either present or they are
not. There is no inherent hierarchy for
the attributes. For example, sex/gender is a nominal measure with two
attributes: male and female. The
attributes are not ranked or privileged in any way; they’re simply qualities
that you either are or are not.
Ordinal
measures are ranked according to some characteristic. For example, “happiness” has attributes that
can be ranked from most to least happy: ‘very happy’, ‘sort of happy’, ‘not too
happy’, and ‘the world sucks’. It’s not
a very precise ranking, but it is nonetheless a ranking. She who is ‘sort of happy’ is clearly having
a better day than he who thinks ‘the world sucks’.
The shortcoming with ordinal
measures is that we don’t know the exact distance between the levels. For example, the difference between ‘very
happy’ and ‘sort of happy’ the same as the distance between ‘not too happy’ and
‘the world sucks’? This brings us to
continuous variables as the distance or difference between attributes is fixed
and can be precisely measured.
Interval
measures are ranked but also tell us how far apart the attributes are on some
scale. Temperature and IQ scores are
common examples. The distance between an
IQ score of 100 and 101 is the same as the distance between a score of 115 and
116. Likewise, the distance between a temperature
of 50 and 60 Fahrenheit is the same the distance between 90-100F. The key point is that interval measures have
no meaningful “zero”—an IQ score of 0 doesn’t mean you have zero intelligence
nor does 0 degrees F mean you have no warmth.
Interval measures are uncommon in sociology though one example you might
come across is time measured in years.
For example, the distance between the years 1980-85 is the same as the
distance between 1995-2000.
Ratio
measures can be ranked, the difference between attributes is fixed, and they
have a meaningful zero. You can be 0
years old and the difference between 10-12 years is the same as the distance
between 20-22 years. Income is another
popular ratio measure.
Finally, I has stressed the
difference between measurement precision and measurement detail. Nominal level measures are considered the
least detailed while ratio measures are considered the most detailed. However, that doesn’t mean that ratio
measures are necessarily more precise.
For example, it doesn’t make sense to say that knowing someone is male
is more precise than knowing that his annual salary is $15K. These are different variables. Precision and detail become more meaningful
when a single variable is measured at different levels. For example, an ordinal measure of annual
income is “a little” while a ratio measure is $15k. The latter is certainly more precise than the
former.
Consider another
example—height. Ordinal measures of
height are: very tall, tall, medium, short, and very short. Ratio level measures are 84 inches, 83
inches, 82 inches, etc. The second
measure is much more precise. Ideally,
we’d always like to have such precision but it in reality this is not always
feasible. More important is that we
understand different levels of measurement so that we can better work with the
data we are presented. And that’s the
moral of this posting!
Key Points: Chapter 6 Measurement
I’m going to talk about two key facets of measurement: 1)
conceptualization and operationalization, and 2) validity and reliability. I cannot stress how important it is that
these concepts are clear in your heads.
One question had to do
with the difference between conceptualization and a conceptual definition.
Conceptualization is the process of developing or “fleshing out” a theoretical
construct by giving it a working definition. The book uses the example of
“prejudice” (108-9). The discussion is thorough so I won’t repeat it here, but
it may help to think of conceptualization as a process that culminates with a
conceptual definition. This “process,” incidentally, doesn’t mean a haphazard
personal definition, but rather doing one’s homework by consulting multiple
sources to produce an informed conceptual definition of a construct that other
people can clearly understand.
In turn, the conceptual
definition informs the next step of the process, operationalization. When
operationalizing, we determine an appropriate method to use in measuring our
original construct. Depending on how specific our conceptual definition and
what we want to know, a researcher could use a survey or field observation or
personal interviews or any number of methods to measure a construct.
Another question had to
do with the difference between conceptual and empirical hypotheses. A
conceptual hypothesis is when a researcher surmises that a relationship exists
between variables, whereas an empirical hypothesis is a definite claim about
how variables are related or influence one another. In the conceptual stage we
think through options of what variables mean and how they are related, while in
the empirical stage we assert a definite claim about how variables interact.
Think of it like betting
on horses at the racetrack (something I’m sure all of you do regularly). First,
you would conceptually review the options: “3-Legged Nag” doesn’t sound very
promising but “Thunderbolt” just screams of a big-money-winner. Next you
observe a few races and indeed find that Nag looses every time while
Thunderbolt consistently places in the top three. Eventually you put money down
on the horse you think will win based on what you’ve studied and reflected on.
You’ve gone from a conceptual process to an empirical venture.
I recommend rereading
page 111, paragraph 1 as it walks you through the conceptual-empirical process.
Reread that paragraph and then actually map out the stages on a piece of paper.
Yes, I’m serious—drawing diagrams is a great way to learn this stuff!
One last question was
about why internal consistency or reliability matters. I guess the most basic
answer is that it only matters if you value reliability. Given the complexity
of social phenomena, we want our measures to be as reliable or dependable as
possible. The text notes that we improve reliability by clearly defining
constructs, using precise levels of measurement, using multiple indicators, and
by pilot testing.
Specifically, multiple
indicators enable us to measure a construct in different ways. Returning to the
text’s example, prejudice does not exist in people’s attitudes and actions in a
single way. Rather, it is manifest in different feelings and behaviors. It is
therefore much more informative if we can measure multiple facets of prejudice
such as attitude, popular belief, ideology, and behavior. In developing
multiple indicators, we increase reliability by measuring more content. It also
helps us root out weaker measures. For example, say that 3 of 4 measures are
highly correlated while 1 is not. It is likely, then, that the lone measure is
either a bad indicator of the construct or that it is somehow erroneous (like
maybe it’s ambiguously worded which leads people to respond erratically instead
of reliably).
Test 1 Review
Here are some review question to help focus your
studying. Bring questions to class. Good luck.
1. From lecture, what distinguishes a research
topic from a research question? A research question from a testable hypothesis? (Hint:
Consider the function of variables)
2. What is meant by
linear and nonlinear paths of research?
What this means for qualitative and quantitative research questions?
3. From the text, why
are variables important in quantitative research? Why might they problematic
for qualitative research? From lecture, what is the difference between the
independent and dependent variables? From tutorial, what’s a trick for
distinguishing between them?
4. In you own words, why
doesn’t qualitative research typically begin with a hypothesis?
5. According to lecture,
most errors in reason are a result of what?
6. Understand the
breakdown of a hypothesis statement. For example: Smoking one or more packs of
cigarettes per week increases risk of lung cancer by 20%. Answer the following questions.
- What are the independent and dependent
variables?
- What kind of relationship is implied?
- What would be the null hypothesis?
- What is the research question?
- From the text and lecture, why are
hypotheses never proven?
- In your own words, what is the "process
of a hypothesis" over time? How might this process apply to the
smoking example?
7. What is the
difference between the ecological fallacy and the error of reductionism? Consider the difference between the follow
statements.
- I’ve seen an awful lot of people in
California driving Hybrid cars. As a result, the California state
government is going to give tax breaks to people who purchase Hybrids.
- Hybrids are now the top selling car in
California. Chris is from
California, therefore he’s probably going to purchase one too.
8. What’s the difference
between an ontological position and an epistemological one? Why does it matter for methodology?
9. What are the
different characteristics associated with qualitative and quantitative types of
research? Although these dichotomies are
more flexible in actual research, why are they a useful way to learn about
research types?
Key Points: Chapters 3 Ethics
This week we had an
ethics marathon. We discussed many
examples including Milgram, Zimbardo, Humphreys, Tuskegee, Olivieri, and
Chandra.
For the exam, you’ll
need to know about each case discussed in the text and during lecture. While memorizing every detail is unnecessary,
you should be able to recognize the details of a specific case even if you’re
not given a name like Milgram or Zimbardo.
For each case you should be able to identify the key ethics issue and
explain why there was a violation. And
if you want to go hog wild, try locating the different cases in the typology of
legal and moral action (p.46).
Remember that there are different kinds of ethics
violations. On the one hand is
scientific misconduct—i.e., types of cheating like fraud and plagiarism. On the other are issues involving research
participants that occur when people are harmed.
Also remember that there are different types of harm that
human subjects can experience in research.
The harm experienced by subjects in the Milgram study is different than
that incurred by subjects in the Tuskegee case.
You will need to be specific on the test.
And finally, realize that we don’t show the films for
entertainment value alone. We expect you
to have viewed and to understand them.
The films demonstrate course themes.
You should be able to identify which theme(s) and then explain how it is
illustrated in the film.
More Practice: Spuriousness
During lecture we discussed spuriousness. Here are more examples. Take some time to think through them...they would make excellent test questions (hint, hint)
Based on the criteria we discussed, are the following examples cases of correlation or causation? If correlation, can you identify spurious influences?
Higher beer prices 'cut gonorrhoea rates' http://news.bbc.co.uk/2/hi/health/729298.stm
Video games 'increase aggression' http://news.bbc.co.uk/2/hi/health/720707.stm
Or, ponder this: Dr. John Harvey Kellogg (co-inventor of the breakfast cereal Corn Flakes in the late 1800s) warned against the dangers of self-abuse including smoking, drinking, non-procreative sex, and (*gasp*) masturbation! In fact, he reasoned that masturbation caused acne. (Some sound reasoning, eh?) Might this be a spurious relationship? Can you think of another variable that might explain the apparent relationship between pimples and self-gratification?
Based on the criteria we discussed, are the following examples cases of correlation or causation? If correlation, can you identify spurious influences?
Higher beer prices 'cut gonorrhoea rates' http://news.bbc.co.uk/2/hi/health/729298.stm
Video games 'increase aggression' http://news.bbc.co.uk/2/hi/health/720707.stm
Or, ponder this: Dr. John Harvey Kellogg (co-inventor of the breakfast cereal Corn Flakes in the late 1800s) warned against the dangers of self-abuse including smoking, drinking, non-procreative sex, and (*gasp*) masturbation! In fact, he reasoned that masturbation caused acne. (Some sound reasoning, eh?) Might this be a spurious relationship? Can you think of another variable that might explain the apparent relationship between pimples and self-gratification?
Refresher: Quantitative Research
On Wednesday, I reviewed the relationship between theory and research. I explained that there are two basic types of social research: quantitative and qualitative. Each is associated with its own ontology, epistemology and direction of theorizing. Several students came to office hours with questions that I want to share for everyone's benefit. Here's the short and sweet of it:
* Quantitative research is associated with realism,
positivism, and deduction.
* Qualitative research is associated with the constructivism,
interpretivism, and induction.
* We worked through an example of quantitative research. We tested our theory of political ideology by
surveying 1000 YU undergrads. Everyone
we surveyed was asked the same question and given the same response options.
Each tutorial established a different “threshold” to confirm or reject the
hypothesis. If the threshold was met and the hypothesis was confirmed, then we
concluded that our theory was supported; however, if the threshold was not met
and we rejected the hypothesis, then we concluded that the theory was not
supported.
* In sum, we began with a theory about neoconservatism and
tried to test it with a hypothesis about voting behavior (deductive). We
assumed that our definition of neoconservatism was shared by our subjects and
would be consistent with their voting behavior (realism). All steps of the
research were predetermined, followed the scientific method, and the process
was objective (positivism). We surveyed 1000 people and collected numeric data
(quantitative).
* This example is meant to illustrate how quantitative
research reflects the particular concepts associated with it. Similarly, you
should understand how qualitative research is reflected in its associated
concepts.
Key Points: Quantitative vs. Qualitative research
Last week I reviewed Ch 2 material on the relationship
between theory and research. I began by
explaining that there were two types of social research: quantitative and
qualitative. Each is associated with its
own ontology, epistemology and direction of theorizing.
Quantitative research is associated with realism,
positivism, and deduction.
Qualitative research is associated with the constructivism,
interpretivism, and induction.
We worked through an example of each. In the first example, we tested our theory of
political ideology by surveying 1000 YU undergrads. Everyone was asked the same question and
given the same response options. In both
tutorials, we had to reject our hypothesis because the survey results were
below the minimum threshold. Our data
did not support the theory we were testing.
All steps were predetermined and the research was objective and could be
easily replicated with similar results.
In the second example, I was running for office and needed
to develop and idea of what issues were important to you as voters. I personally interviewed each of you. We chatted for a bit a what issue was
important to you, why it was important, and what kinds of things I could do if
elected to help address your concerns.
My assistant took interview notes and then as a class we looked at all
the data and tried to develop a broader understanding of what most voters
wanted. Every interview was different
and raised different issues, explanations, and solutions. I learned a lot more about what was important
to all of you, but keep in mind that another group of students might have
voiced entirely different issues and solutions.
Hopefully these examples helped clarify the relationship
between theory and research. And
hopefully it’s clear why the different examples represented specific
ontological and epistemological assumptions and directions of theorizing. Come see me in office hours is you still have
questions.
On a separate note, thank you for humoring me through the
examples. I was elated with the level of
participation by everyone last week!
See you Monday.
Key Points: Chapter 2 Approaches to Research
Here’s more discussion of chapter 2. Normally I don’t go into such detail but time
was short on Wednesday, so I’m spoiling you this week!
First we said that ontology concerns the nature of reality.
Is there a concrete social reality that we all encounter or do we each
experience life as a series of unique, subjective events? We also focused on
two extreme positions: 1) objectivism or realism, which maintains that
we do share a common experience of reality, and 2) contructivism, which
contends that our experiences and understandings of the world are completely
unique to us and only us.
Next we said that epistemology is the study of knowledge.
How can we best learn about something social? We focused on two extreme
epistemological stances: 1) positivism, which advocates that we can
study and learn through the objective application of scientific method, and 2) interpretivism,
which asserts that social life is too complex to study methodically; instead we
must immerse ourselves in research and do research on a case-by-case basis.
Okay, that’s the theory.
Now let’s apply these general concepts to the specific experience of illness
(or racism, homophobia, etc). Some 70,000 people die of cancer every year in
Canada. Length of survival varies, as do types of treatment and care. There are
many kinds of cancer with different causes. Is the experience of this illness
the same or different for everyone?
When the text says that
the ontological and epistemological stances of the researcher matter, this
means that what we believe influences how we research issues.
1. An objectivist
would argue: Although there is no single type of cancer or treatment, and
although the disease attacks people’s bodies differently, as human beings we
still share common experiences of pain, pleasure, sadness, and happiness.
Therefore, people with cancer likely share many common experiences.
2. A constructivist
would argue: Not only do types and treatments of cancer vary, but so do the
ways that people ail and cope with the disease. Therefore the experience of
this illness depends solely on the individual.
3. A positivist
would argue: We can accurately study and learn about collective experience of
having cancer is by asking many people systematic questions. If lots of people
tell us the same (or similar) account, then that is likely a pretty accurate
knowledge of the experience of this illness.
4. An interpretivist
would argue: The way we accurately study and learn about the experience of this
illness is to become deeply immersed in people’s lives. General questions only
tell us so much; in-depth study of representative cases also yield valuable
knowledge. Only this will enable us to
truly understand how people similarly or differently experience this disease.
You’re all entitled to
your own epistemological and ontological positions on these. We care simply
that you understand the different positions and how they influence the way we
methodologically conduct research.
Key Points: Chapter 2 Critical Research Approach
Monday lecture touched
on “critical” research. I want to make sure it’s clear.
According to lecture, the
critical approach is situated closer the interpretivist perspective than the
positivist. And if you think back to Chapter 1, the critical approach is
reflective of applied research (in particular action research, pp. 12-13).
Technically speaking the
critical approach blends multiple epistemological and ontological perspectives.
This really isn’t important since we don’t want you bogged down in
lay-philosophical debates. So, the take-away point is that the critical
approach views knowledge as power, both as a form of oppression and of
empowerment.
Knowledge can be
oppressive when it used to negatively impact people’s lives, yet it can be
empowering if mobilized to improve social conditions. For example, for a long
time many people assumed that homeless people were on the streets because they
were lazy, subject to vice, and made poor decisions. This was common wisdom
and, as a result, the public had little interest in helping the downtrodden.
But once scholars collaborated with social workers to identify and the many
structural factors contributing to homelessness, we began to see a greater
public willingness to allocate resources to mitigate factors influencing
homelessness.
In this example,
knowledge was mobilized as a form of activism for the express purpose of
improving public welfare. Knowledge was sought not for knowledge sake, but
rather in the interests of social reform. This is what the text refers to as
praxis. However, implicit in this approach is the assumption of an objective
reality—that is, that harmful structural conditions truly exist but can
nonetheless be curtailed. There is an implicit assumption of shared
experience—i.e., homelessness—that can be remedied to some degree. At the same
time, few would argue that the experiences of all homeless people are the same
or that people on the streets are all there for the same exact reasons. This
means that the critical approach sometimes straddles that ontological border
between realism and constructivism.
In short, the critical
approach is about producing knowledge for the sake of mobilization and reform.
Whether we employ a positivist or interpretivist approach toward knowledge is
irrelevant; more important is an understanding that power and use of knowledge
to fix social problems.
More Practice: Name that Time Dimension!
For the following scenarios, name if the time
dimension is cross-sectional or longitudinal. If it is longitudinal, is it a
time series, cohort, panel, or case study?
- Last
week a market research company phones up 2,500 people in the GTA to out if
they like Rob Ford.
- A
group of small children all affected by autism are being studied from
onset to their teens.
- Each
year, York gives the same student satisfaction survey to a random sample
of York students. It has been going on for 10 years.
- I’m
going to do a study of 300 households in TO and administer surveys to the
same household every year for the next 15 years.
Come see me in office hours or after class for the answers!
Key Points: Chapter 1
1. Key Course themes
One goal of this course
is that students become informed consumers of information because there is a
lot of "bad research" out there. Learning research methods is an
excellent way to develop a skill set that will help us evaluate information we
encounter on a daily basis.
As social researchers,
we must be especially skeptical of information that is based solely on claims
of authority, tradition, common sense, etc. However, we also need to be
skeptical of information that claims to be science-based as even scientific
research can be poorly conceived, incorrectly done, or presented in a
misleading fashion. Don’t confuse skepticism with cynicism; cynicism implies distrust
and negativity while skepticism refers to evaluating evidence instead of
blindly accepting claims.
According to the
lecture, the scientific method may not be perfect but it is the most accurate
and most consistent technique we have for evaluating the physical world.
“Positivism” is the position that principles of the scientific
method—originally developed for the physical sciences—can be effectively
applied to social research.
2. Key Lecture Points
Know the steps of the
research process (which we will continue to go over), such as selecting a
topic, developing a research question, designing a study, etc.
Understand the
significance of the peer review process. Who qualifies as “peers?”
Understand the
differences between basic and applied research. How are they
differently applied? What are pros and cons for each?
Know the three types of
research: exploratory, descriptive, and explanatory. Why did the lecture
emphasize explanatory research?
Understand differences
between cross-sectional and longitudinal dimensions of time.
Don’t over-think
examples of cohort studies. If you look
close enough, I’m sure that you could find some commonality between the
research subjects. But this isn’t the
point. The key to a cohort study is that
the common quality or characteristic is directly relevant to the research
question.