Elaboration Paradigm compares bivariate to multvariate relationships.
In a Bivariate Relationship you have an IV and DV. A cause and
an Effect.
We introduce a TEST VARIABLE to see whether the relationship is causal.
Test Variable
Partial relationship
compared with Original
Antecedent Intervening
Same relationship Replication Replication
Less or none Explanation Interpretation (explanation is a synonym for "spurious")
Split
Specification Specification
(Split means that one partial is the same or greater, while the other
is less or none).
How do we assess this empirically? We have learned two methods: Crosstabulation (5a). Here one compares the cross-tabulation tables obtained with parts of the sample with those obtained for the whole sample. See Notes of April 7. This method is good for nominal variables, variables with a few categories. The other method we have used is path analysis which was your assignment for today. A third method, which we have not had time to do as an exercise, is to look at interrupted time series data. I will cover this in class although we do not have time to learn how to do the graphs in Excel.
Examples: do storks bring b abies? What is our IV?
What is our DV? What test variable would we introduce?
Does attending an Ivy League College Bring Success in Later Life?
What test variables?
When you have a lot of test variables, path analysis works better than cross-tabulation because cross-tabulation divides things up into so many groups that they become too small. But regression is also a tricky method because the variables are usually interrelated in ways that conflict with its statistical assumptions. You need to test each of your causal arguments separately, do not trust studies that claim to control for everything with one big regression equation. See my paper on Myths of Murder and Multiple Regression. The longer version of this paper used time series analysis to check on the multiple regression modeling. We need to be skeptical of research published in books and journals. Have they successfully proved all the steps in their argument? Did they introduce the right variables? If they use a regression model, can it predict trends? Do the results hold up if you test them with different methods or are they an artifact of a single method?
For your next assignment, I am asking you to look at ONE professional
article published using the same data that you used for your analysis.
See if you can understand what they did. Are you convinced by their
argument? See Library
Assignment.
April 21 & 23: How to prepare a Poster Session and get Extra Credit. Power Point of poster session pages. These pages can be printed and combined with pages produces with Microcase, Word or other software. You can also prepare all the pages in Word. Here is another example of a path analysis assignment.
April 18: How to do the Path Analysis assignment in Word. We did an example in class. The diagram in this example is too small, you could do a better diagram if you made it larger and put it on a longitudinal page (click on "file," "page setup" and "paper size"). To do a drawing in Word, click on "view," "toolbars," and "drawing." The file we did in class is available here.
April 16: We learned how to do path analysis with Microcase.
April 13 - Use of Regression in Causal Modeling. Path Analysis. Examples: Otis Dudley Duncan, "Path Analysis: Sociological Examples." "Occupational Achievement in Australia and the United States: A Comparative Path Analysis."
April 11 - special review session on statistics.
April 7. We worked through the introdution
to exercise 5a in the workbook, and learned how to introduce a control
variable in cross-tabulation analysis. We constructed the following
table based on an example in the workbook:
Sexual Frequency of Divorced/Widowed and Never Respondents
by Age
Under 50
Over 50
Total
Frequency
Of Sex
Div/Wid Never Married Div/Wid
Never Married Div/Wid Never
Married
Less than
29.7% 30.8%
77.9% 70.2%
54.7% 34.0%
Monthly
Monthly or
70.3
69.2%
22.1% 29.8%
45.3% 66.0%
More
N = 313 517 340 47 654 564
Chi sq =
.10
1.393
52.384
P =
.752
.238
p<.0000
April 4
The Art and Science of Cause and Effect. (powerpoint)
Probabilistic cause, not an absolute cause, not a cause
that is sufficient or necessary. "Cigarette smoking causes
cancer." WHat we mean is, smoking cigarettes increases
the likelihood of getting cancer. How much?
There are multiple causes for everything. What we
want to find out is how much each thing contributes. There are also
causal linkages, or indirect causes. A causes B
and then B causes C.
Diagraming causal models. We put the dependent variable
at the right. We draw arrows going into it for each causal
variable that effects it directly. Then we can
have arrows that go into the arrows, steps into the causal analysis, as
in
this sample file:
http://crab.rutgers.edu/~goertzel/homomale.htm
Criteria of Causation - how do we know that something is a cause of something else.
1. Time Order. The cause comes before the
effect. Sometimes we sort out the time order theoretically, we assume
that
education preceeds employment. Or we can use a
research design that involves gathering data at two points in time.
If
you don't have measurements at two points in time, this
is shaky.
2. Correlation. The two variables vary together.
When one is high, the other is high OR when one is low the other is
high. This gets at the degree of causation, the
higher the correlation the strong the causal relationship.
3. non-spuriousness, we want to know that
the correlation is not cause by something else. We can test this
with an
experimental design, if feasible. Or we can use
statistical controls, which are not quite as convincing but its all you
do
in many cases.
We test for non-spuriousness by introducing controls.
Causal Models: representations of the complex causal relationships between variables. Variables have different causal roles, but this is determined by our causal our causal model, it is not inherent in the variables. One person's cause can be another's effect.
Dependent Variable - that is what we want to explain. Often these are opinions or behaviors
Independent Variable - what we use to explain it.
Often there are traits or physical characteristics, e.g., sex or race,
almost always independent.
If you studies the relationship of race on voting, for example, race would be independent and voting dependent.
Antecedent variables, things come before the independent
variable. This helps us to deal with a causal chain.
Antecedent variable cause IV which causes the DV.
If the antecedent variable "explains" the relationship,
we have an "explanation", we say it is "spurious".
Intervening Variables, this that are intervening, e.g.
Race determines ideology which determines the vote.
This is an "interpretation" it tells WHY the causal
relationship exists.
Path
Models: a way of graphically expressing complex causal models.
Example: Determinants of Adult Homosexuality in White Males.
Example: The Seattle
Social Development Project.
Example: Manuscript Pages 6 to 9 of Performance Measures of Effectiveness.
MARCH 24, EXPERIMENTAL METHODS
March 1 -
Experimental Research. Essential characteristics:
March 12 -
Let's look at some of these statistics and how they are used.
The difference between descriptive
and inferential
statistics.
Descriptive statistics we are learning:
Measures of central tendency: mean, median, mode. The concept
of "average"
Measures of dispersion: range, interquartile range, standard
deviation, variance - how much do they vary
Frequency distribution. a way of graphically displaying the central
tendency, dispersion, "shape" - histogram.
Percentages (ratios of a frequency and a base). Univariate percentages
describe frequencies for the entire sample in standardized form, controlling
for sample size. Crosstabulations provide bivariate and multivariate
percentages that measure how strongly two variables are related.
Correlations: measures of the strength of the relationship between
two variables. Includes "Pearson's r" and "Cramer's V" Vary
from 0 to 1 or from -1 to 0 to +1. They measure how well one knowing
one variable helps you to know another
Descriptive statistics describe the sample, they do not reflect
the sample size.
Inferential statistics: - they give you a "p" or "probability"
they measure the likelihood that this relationship occurred by chance.
Or that a difference between two things could have occurred by chance.
Margin of Error for a Mean or a Proportion (we learn to calculate these
by hand). A margin of error tells you how much your results might
be "off" due to sampling error.
margin of error for a correlation.
t-test for difference between two mean scores (used more for experiments,
calculated by computer)
Chi-square test, compares the expected frequency with the observed,
to tell you whether the difference could have occurred by chance.
The inferential statistics depend on sample size, they tell you
how well you can generalize from your sample. it is a good idea to
check them first, to see if you just have random "noise".
Example of a NY Times Poll. Only univariate percentages. A study done by the Methods of Research Class in a previous semester.
Among 78 student in a methods class, 65% could solve this question.
What is the Margin of Error for this percentage?
What percentage couldn't solve it? 35%.
M = 2 * SQRT((p * (1-p))/N)
First, you convert the percentages to proportions, e.g., 65% becomes
.65.
then subtract .65 from 1 to get .35 (1-p)
put .65 in the calculator and hit the * button.
put .35 in the calculator and hit the = button.
hit the/ key and enter 78
hit the = key
the hit the SquareRoot key
Then hit * and enter 2 and hit the = button.
We end up with .108 which we can then convert to 10.8%.
The confidence interval is 65% +/- 10.8%
That goes from: 54.2% to 75.8%
March 10 - The simplest formula is M=1/SQRT(N). This is used when we don't know the percentage results. It gives a general margin of error (based on a 50/50 result). The Margin of error depends only on the sample size in this formula, not the population size. For example with a survey of 1024, M =1/SQRT(1024) = .03125 = 3.12%. This means that if 55% of the sample agree with an item, we can be 95% sure that in the population the true is plus or minus 3.12% from the sample statistic. The 95% Confidence Interval is from 55-3.12% ro 55+3.12%. Or from 51.88% to 58.12%.
How about a sample of 400, what is the margin of error? = 5% margin of error.
How about a sample of 1000. = 3.16%
How about a sample of 10000? = 1%.
65% of a sample of 2000 people oppose war with Iraq. That is our
"P" in the formula. M = 2 * SQRT((p * (1-p))/N). M =
2 (SQRT( .65)*(1-.65)/2000).
M = 2 (SQRT( .65)*(.35)/2000). = .0213 or 2.13%
What if we used Formula One? M = 1/SQRT(2000) = 2.22%
Formula 3 is when you are asked, 'How large a sample is needed."
It depends on the error you can tolerate, not on the population size.
It is N = 1/(M*M)
M = is the margin of error expressed as a proportion. If
you can accept 3% error, then M = 03. N = 1/(.03*.03). = 1111.11
a sample of 1112 is needed to get a 3% margin of error for questions that
are answered by all respondents. N is the number of respondents who
need to answer the question, for a question to be asked only of women,
for example, it is the number of women.
In a sample of 400, the average IQ was 127 with a standard deviation of 5. M = 2 * sd / SQRT(N) / M = 2 * 5/SQRT(400). 10/20 = .5 (not 50%) IQ points. The true figure would be between 126.5 and 127.5 IQ points.
1. In a college class with 125 students, 32 of whom are black, the mean on the midterm was 75. The standard deviation was 8. What is the margin of error for this mean? Formula Four because it is about a mean score. M - 2 * 8/SQRT(125)
2. A researcher wants to obtain a margin of error of no more than 5% in a survey of a county with a population of 300,000. How large a sample is needed? Formula Three because it asks how large a sample is needed. M = .05 N = 1/.05*.05.
3. 55% of the respondents in a survey of a state with seven million Republican voters voted for Bush, 45% for Gore. There were 625 respondents. What is the margin of error for the percent voting for Bush? Formula two because we are given a percentage. M = 2 *SQRT (.55*.45)/625)
4. What is the margin of error for the Gore vote? exactly the same
5. In a survey of 1000 voters, 600 were Democrats, 300 Republicans and
100 Libertarian. 65% of the Republicans favored George Bush in the
primary. What is the margin of error for this percentage? Formula
two M = 2 *SQRT (.65*.35)/300)
6. A survey is to be conducted of attitudes among white, black and hispanic
respondents in Camden County. The population is 300,000. Of this population,
80% is white, 15% is black and 4% is Hispanic. The researcher wants to
achieve a 3% margin of error for the estimates for each of the groups.
How large a sample is needed? Formula 3. N = 1/.03*.03 =
1112 * 3
March 7 - Visiting Lecture by Myra Bluebond-Langner. Hand in a page of notes at the end of the hour to get credit for an assignment.
March 5 - Lecture on Field Methods
The
Review Glossary is not adequate as a guide to this chapter. Some
points to be covered:
Some example of field research:
Margaret Mead, the
only anthropologist (or 
sociologist) to get her own postage stamp, won fame through field work,
primarily her book Coming
of Age in Samoa. Later, this book was denounced by anthropologist
Derek Freeman in his book Margaret
Mead and the Heretic : The Making and Unmaking of an Anthropological Myth.Anthropologists
have come to Mead's defense, and
have restudied the case, but I would have to agree with your text that
"had Mead come back from Samoa with an accurate ethnographic report, it
would not have made her famous." Here is the NY Times Review of Freeman's
critique of Mead.
More recently, there has been a raging controversy about the book Darkness
in El Dorado about research on the Yanomamo in Venezuela is the latest
ethical controversy, which also raises important methodological questions.
Many of the book's allegations, however, have
been contested by the National Academy of Sciences.
The combining
of fiction with factual research is increasingly common both in anthropology
and in biographies. Sometimes this is openly done as a literary form,
in other cases such as that of Rigoberta Menchu, it is only admitted when
critics discover it.
There are many problems
with field research: ethical issues, problems of reliability and
validity when data are gathered by only one researcher, etc. A controversial
book is Laud Humphrey's Tea
Room Trade, which raises ethical issues. He studied gay sex in a men's
room in a park in St. Louis, without informing the participants what he
was doing.
Field researchers sometimes seem to find examples that fit their preconceptions,
and their work is often ignored by those who do not like the results, e.g.,
Leon Dash's book When
Children Want Children and
Rosa Lee which are just ignored by welfare advocates who prefer
more sympathetic treatments. One of the best field studies is
Kathryn Edin's book Making
Ends Meet. which is highly sympathetic to the mothers. However,
Edin collected statistical data as well her illustrative observations.
The statistics showed that almost none of the mothers actually lived off
their grants alone. Eli Anderson's book Streetwise
on men in a Philadelphia ghetto has been well received, in large part because
goes beyond one-sided advocacy.
A great strength of field work is observing behaviors that the people themselves
don't understand or aren't even aware of., or at any event, are unable
or unwilling to talk about. Anthropologist Jules
Henry spent a week living in each of the homes of several children
who had grown up mentally ill, 
trying
to discern patterns in the family interactions that contributed to the
illness. Myra Bluebond-Langner's book The
Private Worlds of Dying Children has been very influential; and
she also has published a sequel called In
the Shadow of Illness : Parents and Siblings of the Chronically Ill Child
She will share her latest research with us tomorrow.
Field reserch offers
a richness of description and possibility of new insights that is unparalled
by any other method. Unless it is supplemented with other methods,
it does not provide statistical data, and it is hard to replicate.
March 3 We will begin the unit on sampling, chapter 4 in the text. The statistical formulas will be covered on Wednesday. The Gallup Organization has a good online presentation on this.
Key concepts:
The size of the sample depends on the amount of error we can tolerate, NOT on the size of the population.
Population and a sample. A sample is selected to represent a population. We want sample members to be typical, this is often guaranteed by random selection.
Often we use stratified sampling, in order to guarantee enough members
of smaller groups in the population. Logically or mathematically,
we are sampling sub-populations. We need as
many individuals from smaller groups as from larger ones if you want
accurate statistics about the smaller groups.
Cluster sampling is done for convenience because we don't have a list
or can't access a total population conveniently. Generally this is
done with geographic areas, "census tracts."
This is done to save time and money. Stratification is done to
get adequate sample sizes for sub-groups.
We also have non-random samples when necessary. "Snowball" sample,
get each person in the sample to recommend others. SLOPS, self-selected
- these are for entertainment. You
can't really generalize to a population.
A stratified sample can be weighted to give accurate figures about the population.
"parameter" - the true value for the population - versus the "statistic"
which is what we got from our sample. The margin of error tells us
how much our statistic is like to be "off" or to
vary from the population paramater.
FEBRUARY 21, we did an in class exercise. Here are the answers:
Descriptive Statistics Exercise
The following are the salaries of officers
in a police department:
The
chief makes $85,000 per year,
Two
sergeants make $60,000 per year,
Six
patrol persons make $40,000 each, and
Two
trainees make $9,000 per year.
1. Make a Frequency Distribution Table for these data. If you do this in class, it can be done by hand. If you submit it to WEBCT, you may type it in ASCII characters or use a drawing program.
Freq Percent
Over 100,000
0 0$
75,000 to 99,999 1
9.1%
50,000 to 74,999 2
18.2%
25,000 to 49,999 6
54.5%
Under 25,000
2 18.2%
Total
11 100.0%
2. Make a Frequency Distribution Histogram
for these data. If you do this in WEBCT, you may create a histogram
with ASCII characters, or do one in a graphics program such as Excel.
Over 100,000
75,000 to 99,999 X
50,000 to 74,999 XX
25,000 to 49,999 XXXXXX
Under 25,000
XX
3. Compute the mean
salary for the department?
85+60+60+40+40+40+40+40+40+9+9 = 463/11 = 42.1
.
60+60+40+40+40+40+40+40+9+9 = 378/10
= 37.8 if the chief resigns.
4. What is the median
salary for the department?
The case in the middle if you put them in rank order. 40
.
9,9,40,40,40,40,40,40,60,60,85 in rank
order.
1 2 3 4 5 6
7 8 9 10 11
The modal income is 40,000. because there are more people in that category than any other.
5. What is the range
of the salaries in the department?
from 9000 to 850000 or 76,000
.
4. What is the standard
deviation of the salaries in the department?
.
(Tronchim
also explains how to compute this, and gives an example.)
X X-bar Difference
Difference Squared
85-42.1 42.9
1840.41 * 1 = 1840.41
60-42.1 17.9
320.41 * 2 = 640.82
40-42.1 -2.1
4.41 *6 = 26.46
9-42.1
-33.1 1095.61 * 2
= 2191.22
4698.91 = the "sum of squares"
divide by n-1 or 10, 469.9 is the variance
sqrt Of 469.9 = 21,7 THOUSAND DOLLARS, NOT %
February 19
Units of Analysis - you cannot necessarily generalize findings about one unit, such as a state or other ecological entity, and another, such as the people who live in it. If you find that states with more money have a higher rate of alcohol consumption, you cannot say that it is the wealthier people in the state who drink. This error is called the "ecological fallacy."
Our main topic today is the Quality of Measures. How do we evaluate and measure the quality?
Reliability means Consistency. Between raters, between testings, between forms of a test, between halves of a test, or between the items of a test. With questionnaires, we measure inter-item and item-total correlations. Cronbach's alpha is a widely used statistical measure of inter-item reliability.
Validity is a much more difficult concept. It asks whether the variable measures the concept it is supposed to measure. This is a philosophical question, what does something really "mean". This is problematic with concepts such as "intelligence" that are unclear in themselves. There are several criteria.
February 17 - campus closed due to snow
February 14 - Happy Valentine's Day! -
To go to the ESS meeting, email BOB WOOD and ask him to put you on
the list, your admission will be paid. wood@camden.rutgers.edu
Measurement - assigning numbers to values in a way that represents information.
What are we measuring? VARIABLES. A characteristic that
varies among the things being studied (unit of analysis). Constants,
things that stay the same.
Nominal measurement, each unit in the study has a unique number.
We could use letters. Catholic, Protestant, Jewish, Muslim, LDS,
Buddhist. We could assign numbers to these, but can't do calculations
with those numbers. Each case>
A sub-category of nominal measurement is dichotomous measurement, where there are only two categories. You can make anything into a dichotomy, e.g., religion. Catholic or NOT; Protestant or NOT. Dummy variables, things that are artificially dichotomized. Computers work that way.
ORDINAL - start with categories and put them in a meaningful order. Rank in class is ordinal. Do you allow ties or not? Letter grades are ordinal. This level of measurement does not permit addition or multiplication, so we can't use descriptive statistics that require adding or dividing, such as the "average". We do percentages.
INTERVAL - We measure the distance between the categories. Height: very short, short, medium or kinda tall, pretty tall, very tall (ordinal measurement). 5'7" 6'2". Interval measurement. I could compute the average and say the average height in the class is 5'9". Test scores are used as interval measures, although one question the validity. Temperature. cold, very cold, warm, hot. 19 degrees F. This winter the average high temperature was 32 degrees.
RATIO - interval measurement with a meaningful zero. Zero is the absence of a phenomenon. This means you can multiply. A has 100 dollars and B has 50 dollars. We can say, A has twice as much money. Today is 30 degrees, monday was 15 degrees. Is today twice as hot? Zero Fahreneheit is not the abscence of temperature.
Feb 12 - We will go over the computation of row, colum and total
percentages
in cross tabulation tables. This is best understood from the in
class exercise. Percentages are usually based on the independent
variable. We normally put the independent variable in the columns,
and present tables with only the column percentages. Example:
a study of attitudes
towards sprawl in South Jersey. Sometimes all
three are presented but I think this is confusing. Sometimes
tables present the actual observed frequencies instead of percentages,
e.g., the NY
Times study of clerical sexual abuse (I missed the link in class before,
it's on the right in a box labeled "multimedia"). The expected frequencies
help us to compare the observed frequencies with what would be expected
under a null hypothesis, e.g., the
study of alleged racial profiling by police officers.
Methods and Techniques of Social Research.
Exercise to be done in class or submitted to WEBCT. Name . .
Consider the following answers to the question "I believe the Canadian figure skaters should have received the gold medal":
85 men agreed
Men Women
95 women agreed
15 men disagreed
Agree
25 women disagreed
Total
Answer the following questions:
1. What percent of the men agreed?85/100 =
85%
2. What percent of the women disagreed? 25/120
* 100 = 20.8%
3. What percent of those who agreed were men? 85/180
*100 = 47.2%
4. What percent of those who disagreed were women? 25/40*100=
25/40 = 62.5%
5. What percent of the respondents agreed? 180/220*100
= 81.8%
6. What percent of the respondents were women? 120/220*100
= 54.5%
7. Fill in the Table:
Gender and Belief that the Canadians Should Have Received the Gold Medal
| Men | Women | Total | |
| Agree | 85 | 95 | 180 |
| Disagree | 15 | 25 | 40 |
| . | 100% (N=100 ) | 100%
(N =120) |
100%
(N= 220) |
8. Now, try figuring out some expected (null
hypothesis) frequencies. What would you expect to be the cell
frequencies if there was no difference between Men and Women on the issue,
given the marginal frequencies provided?
New Example
| Men | Women | Total | |
| Agree | 100*180/220=
81.81 = expected frequency |
180*120/220=
98.18 |
180 |
| Disagree | 100*40/220=
18.18 |
120*40/220=
21.82 |
40 |
| . | 100 | 120 | 220 |
Cramer's V is a measure of the Strength of the
Relationship, not of statistical significance.
We use "chi square" as the statistic to measure
statistical significance for categorical data.
The "p" value is the probability that this relationship could have occurred by change. In this case p = .264. If p is larger than .05, we say "this difference is NOT statistically significant" as in this case.
9. Now compute the Chi Square Statistic, using the WEB Chi Square Calculator. (If you do this in class, the calculation will be done on the screen). Enter the following information:
. 1 .Degrees of Freedom
.
1.248 .Chi Square
. .264 .Significance
(yes or no, or p = ) p > .05
The asterisks.
One * means p < .05 or the probability of this occurring by chance is less than 5 in 100.
Two ** means p < .01 or the probability of this occurring by chance is less than 1 in 100.
Descriptive statistics describe the sample. They tell us how strong the relationships are in the sample (among other things). V and r are descriptive statistics. Percentages are descripitive statistics.
Inferential statistics tell us whether we can generalize from our sample to a larger population. We use chi-square (and other statistics) to calculate a "P" which is the probability that the relationship could have occurred by chance.
We want p to be small, as small as possible.
We want r and V and other descriptive statistics
to be large.
Feb 10 - Research process using survey data.
Gender and Opinion on the Iraq War
Male Female Total
War in Iraq? Yes 7 12 19
No 10 22 32
17 34 51
Yes 41.1% 35.3% 37.3%
No 58.5% 64.7% 62.7%
100% 100% 100%
What percent of that? You ask, percent of what?
What percent of the men agree the government should help? 378/835
What percent of those who think the government should help are men?378/896
What percent of the respondents are men who think the government should help? 378/1849
Under the null hypothesis that gender is unrelated to opinion on this question, how many men would we expect to favor the government helping. Since 48.5% of the sample support it, we would expect 48.5% of the men to support it. .485 is the proportion who agree, multiply that by the number of men. We would expect .485 * 835 = 405
What percent favor the war? Number favoring the war and divide it by the total number of students 19/51 37.3%
What percent oppose the war? 32/51 = 62.7%
What percent of the male students supported the war? 7 males support the war out of 17 male students, 7/17 = 41%
What percent of the students who support the war are male? 19 students supported the war, 7 of them are male 7/19 = 36.8%
Normally we base out statistics on the Independent Variable or the "cause".
February 7 - campus was closed, snow day
Feb ruary 5 - Today we will view part of a CBS TV program on
"Junk Science." Scientific research often is used to advocate for
public policy positions. Unfortunately, this research is often of
dubious validity, and scientists can be found to defend any position.
This video discusses some of these cases. There are some links to
WEB sites that are critical of the video.
My son and I did a biography of Linus Pauling that discussed his research on Vitamin C, which he took up at age 65. To a great extent, his confusion came because of differences between medical research and the chemical research he did during his earlier career. Much of the research on Vitamin C used experimental designs and found small or no statistically significant differences. Pauling often felt that it was up to the researchers to disprove his theories (i.e. to prove the null hypothesis). He also criticized the way experiments were done, saying that if the doses had been larger or if they had been given earlier or to patients who had not had chemotherapy, then the Vitamin C would have worked. New theories of this sort are coming out all the time, e.g., Vitamin C as the cure for heart disease.
February 3 -
Accepting or rejecting hypotheses. If we hypothesize that a relationship exists and is positive, it must be both positive and statistically significant for us to accept it. A hypothesis might also be more complexi, e.g., we might hypothesize that a relationship between two variables will be significant even when we control for a third variable, e.g., that black people will be more likely to vote Democratic even controlling for income. In that case, we would have to show that the relationship holds up under control to accept our hypothesis. Tests of statistical significance allow us to reject the "null hypothesis" that there is no relationship beyond random chance.
Research Design. How research is organized or structured to accomplish different ends.
Experimental Design.
To test causal hypotheses may be the purpose of a study. An experiment is the most rigorous way to test a hypothesis. In an experiment we manipulate the independent variable. The independent variable is the "cause" . Then we measure the dependent variable or "effect" both before and after.
Classic Experimental design
Before
After
Experimental group
measure DV Apply IV measure DV
Control Gp Measure DV Measure DV
Hypothesis: The two groups will be the same "before" and different
"after"
Dependent variable is your outcome - health, getting over your cold,
recidivism, attitude change, etc.
The best way to make sure the groups are the same is to assign people
at random. In a small experiment you may match them on characteristics
as well.
We control for extraneous variables by randomization or by standardization
- making everything the same.
Placebo effect - you feel better because you took a pill and it made you think you would get better.
Some things cannot be manipulated, e.g., race, class, etc. It may be unethical to experiment. Experiment is artificial, the results may not generalize to the real world. "External validity" may be a problem, the ability to generalize.
SURVEY RESEARCH. Best for describing large populations. More for description than testing hypotheses. Questionnaires or interviews. Often on the telephone. Good for things that people can tell you and are willing to tell you. Consumer behavior. Sampling design, make sure the sample is representative. Do cross-tabulations, see how two variables relate. Look at segments of the populatione, e.g., young and old, male and female, black and white.
FIELD RESEARCH. Go into the natural setting and observe what is
going on. You may talk to people and ask questions as well, but the
really unique aspect is observation. Access is a big problem.
Anthropologists do mostly field research. Takes a long time, you
have to overcome people's tendency to present a good image. Covert
vs. overt. Does you being there effect their behavior? Tends
to be INDUCTIVE, get new ideas. "grounded research" means you collect
your observatins, then generalize from them.
Experiments are DEDUCTIVE, you have an idea and then test it.
USING OFFICIAL STATISTICS - Using data that is already being collected by a country, a government office, a company. Criminal justice systems generate a lot of data for their own purposes. You are limited to the questions someone else designed and asked. Data banks of surveys are available, analyze somebody else's survey data.
CONTENT ANALYSIS - Looking at media as a source of data: tv shows, letters to the editor, newspaper articles. Written documents. You can go back in time.
FOCUS GROUPS - Group interviews lasting about an hour and a half. Inductive purposes, collective ideas you didn't have before.
Jan 31 -
Aggregate data is about groups, not individuals. In this case the groups are states of the US.
Correlation: two things vary together. They can vary directly or inversely (positive or negative)
We can see that reading field and stream is positively correlated with hunting rates in a state, but that does not prove that the hunters are the ones reading the magazine. The "ecological fallacy" is to generalize from aggregate data to individuals.
The Independent Variable determines the Dependent Variable, it is usually one of the "causes" of the Dependent Variable. There are several independent variables for each dependent variable in an analysis. Your hypothesis determines what variable is dependent. The dependent variable is whatever you want to explain.
The correlation coefficient is a measure of how much the two variables vary together. A positive coefficient means they vary positively, a negative one means they vary negatively. The correlation measures the strength of the relationship. The asterisk tells you that the relationship probably did not occur by chance with this size of a sample, it measures the statistical significance. Withone asterisk, there is less than a 5 in 100 chance of random error. Two asterisks means a less than 1 in 100 chance.
The correlation matrix summarizes the strength of the relationships between a set of variables. The strength of the correlation is the absolute value, how far it is from 0 in either the positive or the negative direction.
Dductive research: formulate a hypothesis and test it.
Inductive research: look at the data and formulate a hypothesis.
January 29 -
Selecting a topic. Your boss told you to - "applied research." We may get to choose our own topic. Often this is done for advocacy purposes, we want to prove our point to someone else. Come up with data that demonstrate what you believe to be true. You may want to find out something you don't know. Get information you can use in some way. Needs assessment, what services are needed by our clientele, our population. Find out about sexual abuse by Catholic Clergy.
Purposes may be purely descriptive or they may be analytical, designed to test an idea.
Formulate the research question. How many priests are abusing children? How widespread is this issue? When did emerge? How many priests have been accused, how many people claim to have been victimized?
Defining the concepts? What is abuse? Is any sexual contact between a priest and a parishoner abuse, or only with a minor?
Operationalizing the concepts. Writing the questionnaire. The exact questions thqt you are going to task.
Hypothesis? An idea that you are trying to test. Not all research is designed to test a theory, or to test a pre-existing idea. This research is primarily descriptive. Hypothesis: Celibacy increases abuse. Have to have a comparison to test this hypothesis, e.g., Episcopal vs Roman Catholic priests.
Make the observations. Collecting the data. Sometimes we observe behavior, participate in a group and see what is going on. Tearoom Trade, Humphries served as lookout, wrote down the license numbers of the cars. Went to peole's houses and did a survey about their attitudes, employment, etc. Not about their sex life. In most cases, we just ask people questions. On the telephone, in person or even by mail, on the internet.
Analyzing the data - Compile statistics. Frequencies and percentages. Often that is all that we are interested in. Next stage usually is crosstabulation, looking at two variables at the same time, how they related. Sometimes we get data on trends, how things change over time.
January 27 -
What is science? How does it differ from the humanities, from art? Do the social science differ from the natural sciences?
Science attempts to make generalizations. History or art, you are just interested in specifics. In criminal investigation: Who committed this particular crime - not scientific question. What determines the murder rate in society, that is a social science question.
Findings should be replicable, two social scientists should get the same result if they ask the same question.
Scientific knowledge - always true, invariant, true for everyone.
Technical or practical - how to do something, pragmatic.
Ethical - what should we do? What is right?
Scientific knowledge is Empirically Verifiable - tested by objective evidence.
"Jesus Never Fails".
Make statements using CONCEPTS. A concept is a word, or the idea behind a word. "Mother". Madre. Make them explicit. Adoptive vs. birth vs. surrogate. Male or female. Race. Concepts are not objectively true, they don't come from God. In science, we define the concepts to be measurable. The goal is to capture reality. Parsimonious, efficient. Useful for our purposes, to predict or explain, or to change something. Clear boundaries or arbitrarily defined. Draw arbitrary lines, e.g., 18 years old is an adult.
Theories: An abstract statement about the relationship between concepts. Falsifiable, can be proven wrong. We never prove that anything is true, we simply say that it has held up against attempts to disprove it.
NULL Hypothesis. The hypothesis that our statement is wrong. Try to reject the null hypothesis. Women like soup more than men: a hypothesis. Null hypotheis: there is no difference between men and women in preferences for soup.
Men Women
Like Soup 311.11 488.88 800 88.8%
Don't Like Soup 38.89 61.11 100 11.1%
350 550 900 99.999%
On the null hypothesis that there is no difference between men and women on this issue, how many men would we expect to like soup?
What percent of the respondents like soup? 800/900 88.8%, the probability of an individual liking soup is .8888888
The probability that a man will like soup, under our null hypothesis, is .8888. There are 350 men, so we would "expect " that .88888 * 350 or 311.11 men would like soup/ This is the expected frequency.
The numbers in this table are expected frequencies based on the null hypothesis that there is no difference between men and women in their likeing for soup.
Tne number of men who would not like soup would be .11111*350