COMM550/Class Notes

From Driscollwiki

Jump to: navigation, search

01 Sep 09

theory cluster discussion

interpersonal

"one cannot not communicate"

palo alto school, psych-therap
can't engage w/o def of communication
palo: any act in which someone engages is communicative behavior
- must it be interpersonal?
- must there be an audience?
counter: whenev information is encoded and exchanged in a system of symbols
- one can not communicate
symbol - a referent, offered and exchanged

communication relational

must be flowing between two or more ppl

org comm

tending toward qualitative in practice

org was very rational in 40s, 50s, 60s

satisficing model

bounded rationality
same person goes through models until they find one that is satisfactory, good enough and proceed

oscillation between rational, less rational

now an effort to integrate, and balance the 2

mass media

what is 'mass' media?

tied to tv, radio? conventionally broadcast

internet / radio, confusing tech and medium

radio as tech
radio as conceptual collection
internet as tech
but frustrating to consider internet in this fashion, no industrial convention

when geeks think about internet they may say:

there is only one internet

yet when people experience internet they may say:

everything on the internet is x
but this is the experience of a specific collection of internet activities

"internet is dead all i use is twitter."

http://twitter.com/davewiner/statuses/3655166887
invalid comparison?

comm and info tech

contextual design

user-centered design

Communication processes

Health Comm

Language Theories and Linguistics

Model of text comprehension
- Symbology

Media, Culture, and Society

gatekeeper role, professional, informal
- liasons
cultivation
- longitudinal accumulation of effects
- positive ethical implications

Public relations, advertising, marketing, consumer behavior

Semiotics
- mediated representations of things

Footnotes on comm theory

Obvious, common sense

contrary to davis 'interesting'ness,

valuable to prove obvious assumptions empirically

Theory / Model / Hypothesis

Aristotelian approach, testing conclusion
- deductive hypotheses
- inductive inferences
Model-based approach, test all parts

Aristotelian rhetoric

ethos
logos
pathos

Dual mindedness

Nearly all ASC faculty have dual appointment
Publish in journals of multiple fields
Conversant in multiple traditions
- monge first pub == 1973 cybernetics, system theory

Theories in general

Maintain critical view
Remain sympathetic
Track research literature, not theory alone

Experience of theory

Rollercoaster, riding the excitement
Feeling the vibe
Rise, halcyon, decline

Strong

Systematic
Generalizable
Testable
Useful

Weak

Follow up assignment

read literature on a theory of our choosing
summary of the literature, what is said, been said

Research idea development - From your selected communication theories, choose one topic and review relevant literature published in the past 6-8 years in major communication journals. In your literature review, pay special attention to what knowledge claims are made, how they are supported or not, and what the current state of knowledge is in this area. Write a 3-4 page report that summarizes the major findings and identifies the additional topics for future research. Also attach a list of relevant references .

Sept 8, 2009

Lee, "Presence, Explicated"

Step 0, What is the problem?

Concept of 'presence' is confusing, misleading, ineffective for how it might be used in comm research
How does he approach this project?

Step 1, What's been said?

Citing existing definitions
Corpus of existing work
Attempts to draw up history of each definition, theoretical roots
State of affairs (2003) as current as possible
How are they similar, different?

Step 2, Critique, states dissatisfaction

Ambiguities:
- Definitions overlap each other, unclear bounds
Dependence on technology
Question of mediation?

Step 3, Offers new definition

Three parts:
- Physical, sensory based encounters
- Social, encounter w/ other actors
- Self, experience representation of self

Struggling to keep the parts distinct

Constructs, theory

i need this clarified further...

Statement linking two constructs is a simple proposition
Theory requires many such statements
Hypotheses may be built from these observations
But the hypothesis must be testable

Conceptual definition

Operational definition

Defined by operations undertaken to define it
- tautalogical?
- product of various elements you engage in
Around since use in phys sci of the 20s
Recipe-like
- "A cake is the result of mixing flour, sugar, eggs; pouring in a pan; baking; etc."
e.g. Intelligence is what an intelligence test measures

Single Variable behavior

Over time,
- Discontinuous: Is the YouTube up?
- Continuous: How many views might it have?
If magnitude doesn't change, there is no trend.

Two Variable behavior

Two vertical axes
- Same horizontal axis; time
- Different scales on vertical axes; appro to variables
Lag, time after change in indep var has effect on depend var

Causal v Correlation

Correlation, two vars change together
Causal, you or something else is effecting a change in another variable
- Testing an intervention

Experimental design

How is your variable behaving today?

Measurement

Assigning numbers to properties of a phenomenon
If measurement is accurate, one can build a mathematical model and manip the numbers instead of the social system.
- Easier to manip numbers than social sys.

Properties, attributes

Categories, distinct groupings
Order,
Magnitude
Origin

Relationships

Binary (dyadic), 2 vars together
Tertiary (triadic),
Direcitonality

Measurement error

Measured score = true score + error
- True score = Measured score - error
True score is "latent"
Measured score is "observed"

Alt number systems

Dec
Hex
Bin
Duodecimal

Properties of number systems

All of these are basic requirements for measurement and statistical analysis:

Categories (enumeration, distinction)
Order (rank)
Magnitude (amount, quantities)
Origin (true zero)

Researchers assign the numbers. Mathematical system is not sentient.

Sampling in measurement

Domain of measurement
Population of items to represent the domain
Random sample of items
How well does the sample represent the domain?

Characteristics of measurement

What is validity?
- Are you measuring what you want to measure?
What is reliability?
- Consistency

Semantic differentials

Bipolar adjectives

Osgood, attempting to measure meaning
- Good-Bad
- Strong-Weak
- Fast-Slow
- Evaluated using ILLIAC,
  - http://en.wikipedia.org/wiki/ILLIAC

Language intensity scales

- "My supervisor listens carefully to what I say."
- "NO!! No. No? Yes? Yes. YES!!!"

Graphic scales

- :(, :|, :)

Thurstone Scales

Obtain large sets of statements (>100) toward an object, concept representing diff degrees of posi/negi sentiment (attitudes)
Have fairly large num of judges eval statements in terms of favorable/unfavorable-ness, sort into 11 piles labeled A thru K
Assign numberical values to statements from the numbers associated w/ each pile into which the judges place them.
Compute the median and interquartile range for each statement
Select statements that fit best into each numerical entry. Demonstrates near consensus.
Later, reponses are given numbers based on those consensual values

Likert Scales

Strongly disagree, disagree, neither, agree, strongly agree
Imbalanced within itself.
- ie, my strongly disagree may be less strong than my strongly agree

Guttman Scale: Bogardus' Measure of Social Distance

Sometimes called "threshold scales"
e.g. How willing woud you be to admit Armenians:
- To close kinship by marriage
- To your social club as personal chums
- To your street as neighbors
- To employment in your occupation in your country
- To citizenship in your country
- As visitors to your country
- Would exclude from your country
Includes a hierarchy, measure from the highest response
Assumes a particular linearity to the relationship.

Measurement Takeaways

We measure so we can create a manipulable numerical model
Want to capture the props of a phenom via the props of the number system
Many diff kinds of measurement
Measurement results: Observed score, true score, error

Distributions

Distribution: a collection of measurements viewed in terms of frequency of each category

Types of distributions

Uniform
Bimodal, 2 modes
Sinusoidal, oscillating
Power, exponential/logarithmic

Normal, Guassian distribution

Reflected in many natural world phenomenon
Used in comp graphics: http://en.wikipedia.org/wiki/Gaussian_blur

Central tendency

Measures of central tendency indicate where the distribution is anchored.
Mode: most freq score
Median: middle score in the distribution
Mean: sum of the scores divided by the num of them.
In a Gaussian distribution, all three measures are the same.
If they are diff, you can infer a lot about the distribution.

Dispersion

Range: highest score minus the lowest score
Interquartile range: middle half of the distribution
Variance: mean of the squared deviation scores, s^2
Standard deviation: the square root of the mean of the squared deviation scores. Also called root mean squared deviation: s
s = root of s^2 // techincally +/- but use only +

Skewness

+/-

Kurtosis

Flatness or peakness
Leptocurtic, squeezed
Platycurtic, smooshed

Takeaway

Four measures of a distribution:

Central tendency
Dispersion
Skewness
Kurtosis

Lab #3 Descriptives in SPSS

Levels of measurement

Nominal (e.g. "male:1, female:2"), mode
Ordinal (e.g. results of a race.), mode, median
Interval (consistent value for numbers. e.g. temperature.), mode, median, mean
Ratio (e.g. ), mode, median, mean

Style guide

See: sample manuscript in Chp2 of "Publication Manual"
First drafts must be in this style
Send copy of research idea development to monge@usc.edu

Sep 15, 2009

= Purpose of design: controlling variance

Three purposes:

Maximize systematic variance

NO variance == constants
non-systematic variance == random
systematic == variance that you can control

Power: ability to detect an effect that actually exists
- Large variability, power, need only small sample to detect
- Small variability, power need large large sample to detect

Control extraneous variance

Not explicitly linked to your hypothesis
Hold potential independent variables constant
Randomize
- Not always possible
- In the field, you take groups as they are
Make extraneous variables IV to control
Match participants

Minimize error variance

Reduce measurement error
Increase reliability

Designs and Design Criteria

Elements of Design

X = IV, manipulated
(X) = IV, non-manip
~X = manipulable but not
Y = DV
- Yb = DV before manip
- Ya = DV after manip
R = Randomized
M = Matched

Defective Designs

One Group Design

GR1: X Y Exp
GR1: (X) Y NonExp

One Group, Before-After (Pretest-posttest)

GR1: Yb X Ya Exp
GR1: Yb (X) Ya Exp

Simulated Before-After

GR1: _ X Ya
GR2: Yb _ _

Two groups, No Control

GR1: X Y Exp
GR2: ~X ~Y Exp

Threats to Internal Validity

Campbell & Stanley (1963)

Measurement

- infl of measuring instrument on the measured person
- there are "non-obtrusive measures", eugene webb

History
Maturation
Statistical Regression

- Extreme scores on a second test will be less extreme
- Regression toward the mean

Instrumentation

- e.g. Human observers change accuracy overtime
- Instrument goes out of alignment

Selection

- Biased grouping

Attrition

- People drop out

Interaction of otherseven threats

Criteria for Good Designs

Does the design match the hypotheses or answer the research question? (An invariant transformation perhaps?)

Does the design control for Extraneous Independent (OTHER) variables?
- "It's not location, it's randomize!"

Is it Generalizable?** Does the design control for threats to internal and external validity?

Ethics, IRB

Basics

Your study has to copmly with some basic ethical principles for conducting research if it involves human subjects.

If you want to publish, you have to aquire IRB before collecting data, recruiting subjs

NOTE: exempt is still a review status

Ethics

3 Major Ethical Concerns, Belmont Report 1977

Respect for persons
Beneficence
- Don't do the study if it doesn't benefit society
Justice

Implementation of "respect for persons"

Informed consent
Information
Comprehension
Voluntariness
- No coercion

Implementation of "beneficence"

nature and scope of risks and benefits
systematic assessment of risks and benefits
assessment of the justifiability of research

Implementation of "justice"

Individual justice
Social justice
Potentially improve external validity of your research

Statistics and Parameters

Statistics: a characteristic of a sample
Parameter: characteristic of population
Statistical inference: making inference to whole based on some sample

Sampling Distribution

Measuring sample:
- Mean, median, mode
- Range, variance, std dev

Take a random sample (may be representative) from a population
Find mean for that sample
Take new random sample, find new mean
Plot these means on a graph

22 Sept 09

Papers

Send to Prof Monge directly.
Audience of journal readers
Historical account is not necessary
Assume that people who are interested in your area have already read the relevant material
Primary focus: Reporting empirical results
- Audience expecting a certain form for such an article
Literature review section
- Must be contemporary
- Snapshot of recent use of a theory
Citing an article explains connect to your argument
Conclusions
- Hypothesis supported, argument correct
- Hypothesis unsupported, argument not correct
Remainder of article,
- Explanation of success/failure
- Suggest further work in this area
What if another piece of research contradicts our hypothesis?
- Explain weakness in counterevidence experiment
- Synthesize the experiment w current work
- Explain the counterevidence within the limits of the current work
How many articles to review?
- Even in a BRAND NEW area, 10 is too few.
Deviating from APA manual "sticks out like a sore thumb"
- Amateurism

Summary

Use findings from prior work
Use theory
Make an argument
Arguing for positions

Example: Miller on Psychological reactance theory

Arguing for a return to a theory that was abandoned
PRT explained first in the 60s
The paper reviews the earlier work (60s) and recent work (00s)
- An exception to the standard becase they were arguing for a reup

Sampling and probability distributions

You want to be able to make inferences from one sample to the population from which it was drawn
How do you make the right inference?
- You need a sampling distribution of the same size as your sample
Sampling distributions
- In the back of the Hayes book

Class age examples

Data: 23, 24, 24, 25, 25, 25, 27, 27, 28, 28, 28, 29, 29, 29, 30, 30, 30, 31, 32
Freq: 23:1, 24:2, 25:3, ...
Prob: 23:1/19==0.0526, ...

Hypotheses

Null hypothesis
Research hypothesis
- Reject a null hypothesis to accept a research hypothesis
- Andrew's Proof
- Never claim that you prove your hypothesis is true
- Indicate that your hypothesis is supported.
Direction and nondirectional hypothesis
- Null: mu/i = mu/ii
- : mu/i != mu/ii
- Directional: mu/i > mu/ii
- Directional: mu/i < mu/ii
- Where mu/i and mu/ii are hypotheses
Significance level (region of rejection)
- $p < .05$ // willing to be wrong 5% of the time
- $p < .01$
- $p < .001$

Type I & II Errors and Powers of Tests

Type I Error (alpha)

Rejecting the null hypothesis when it is true
Researcher determines the alpha error
e.g., if it falls in 0.05

Type II Error (beta)

Accepting the null hypothesis when it is false

Power

Area in which your are able to detect a particular finding
Possible to computer power of a test in advance
Required by some journals for all statistical tests

Summary

alpha
p value
power of test for given sample size
SPSS will do this

Sept 29, 2009

t-test, z-test

Variance

Total variance == Systemic variance + Error variance
Ration of a statistical test: (syst var/error var)

Systematic variance: variance generated by the experiment. Variance in the dependent variable.
Error variance: sampling error. Any other variable beside what interests us.

Why is Systemic the numerator?
- (1-(SysVar/ErrVar))

If Error var is larger than System Var, no research validity.
No systematic variance, all error variance
- system is undefined

Rule of thumb
- If ratio is smaller than 1, it's a red flag
- Rare to have error var larger than system var.

Variance ratio
- mean of zero
- measured in std deviation units

t distribution

"student's t distribution"

Variance
- mean squared deviations
Std Dev
- root mean squared deviation

Refer to t tables in the back of the hayes text

APA style notes

ethical compliance
chap 2 manual structure and content

Schrock on Walther

Needed a manipulation check

Li, hollingshead, transactive memory

Oct 13,

Consent form
Survey materials
Manipulation
Hypotheses, expectations
Justification, how it will contribute

$Y = b / o + b / 1 X / 1 + b / 2 X / 2 + ... + Error$

October 20, Nonparametric Methods

Recommended journal: Communication methods and measures`

Parametric

Distributional assumptions of parametric tests:

Normality
- No skew, no kurtosis
Homogeneity of Variance
- Sub-groups would have roughly equal variance
Continuity and Equal Intervals
- Dependent variable is measured at interval or ratio level
Independence of Observations
- Each observation has equal opportunity to be selected as part of sample

Violations of assumptions and robustness

How extensive?
- Every assumption is probably violated in some way (except maybe Independence)
Robustness: to what degree does the violation threaten validity
How serious? Many tests are robust.

Remember: statistical tests do not know what the numbers mean

researcher responsibility to assure appropriate measurement

Nonparametric

Nonparametric tests are often called "distribution free"

Don't make assumptions about the parameter distribution
- Normality, Homogeneity of Variance, Continuity and Equal Intervals, Independence of Observations

Single Sample Chi-square Test for Categorical Data

$\chi^{2}= \sum[\frac{(O-T)^{2}}{T}]$

To use result, go to $χ 2$ and locate probability.

Among Sample Chi-square Test for Categorical Data

For multiple groups with multiple levels, we must sum the columns and locate the grand sum

$\chi^{2}= \sum[\frac{(O-T)^{2}}{T}]$

The result is meaningless on its own. Must reference a $χ 2$ distribution table to find $p$ -value.

Tests for ordinal data

Central tendency maps to scale:

Mode: nominal, categorical
Median: ordinal
Mean: interval

Therefore these tests rely on the Median

The Median Test for Independent Samples Ordinal Data

Another nonparametric test useful for ordinal data.

The Mann-Whitney U Test for Independent Samples Ordinal Data

Used frequently in early mass comm
Implements median

The Kolmogorov-Smirnov Test for Independent Samples Ordinal Data

Nonparametric Tests Demo

See examples in C's PPT and SPSS book
Hayes p250

Linear Regression, Nov 3 2009

Reviewing correlation

$r x y$ : correlation, co-variant
$r 2$ :
$R$ : multiple correlation
$R z . x y$ : $z$ varies with $x y$
$R x y . z$ : $x y$ varies with $z$ (control)

ANOVA: compares groups, e.g.

IV: men, women (gender)
DV: n (height)

Correlation: continuous measures, can use more fine grained data

IV: age, in groups 0-1,1-2,2-3,etc
DV: height, 0-4,4-5,5-6,6-7,7+

Important ideas in regression

There is no actual causation, one can always find an alternative explanation

Implied by hypothesis
Experimental design is important to be able to attribute cause
- True experiment is best at establishing causal relationship
- Quasi experiment better than cross-sectional survey

Standard error of estimate

Standard error of estimate (SEE): Std Dev of error, of residuals

Range around the regression line
Really good regression will have as small as possible SEE

Simple bivariate regression

One IV, one DV

Multiple regression

Standardized v. Unstandardized coefficients

Unstandardized is measured using original units (e.g. inches, and minutes) (b)
- Calculate predictor values
Standardized coefficients are necessary for comparison of diff IVs (Beta)

Z-score: normalize a variable with mean of 0 and unit of measurement as std devs.

Notation: $Z gender$

Dummy coding: transforming non-continuous variables (categorical, nominal), e.g.

example 1, nominal:
- we have a Gender variable coded as such {m:1, f:2}
- dummy coding: {m:0, f:1}
example 2, categorical:
- we have Age variable coded according to age groups {0-10,11-20,21-30}
- we create dummy variables for each group {0-10:1,11-20:1,21-30:1}
  - for a given case, if age is within a group, we score it 1.

Lab notes

Bivariate linear regression

Better linear regression models have small std error estimate.

Multiple linear regression

Test relationship among multiple IV and one DV

Important to note difference between significance and Beta

November 17

Presentation prep

December 1
930 - 1pm
10 minutes
"Advertisement" for the paper
Strong "takeaway"
General theoretical background, no thorough history
Need not state hypotheses formally
Focus people on the essential research hypotheses
Results, restate hypothesis, "we tested #1, here's the result, it was/wasn't supported."
Implications, reflection: fatal flaw? Cast doubt on the theory?
Try to leave 2 mins for Q&A
Final slide with contact information
Make copy of paper, presentation available to everyone else in the class

Final exam

December 8
Similar to midterm
Cummulative
Weighted toward the 2nd half (25/75%)
- Non parametric stats
- Item analysis
- Validity, reliability
- Correlation, zero-/first-/second-order, partialing, control for contamination
- Simple bi-variate regression, multiple regression
- Factor analysis
- Time series analysis
- Every statistical test is a ratio between systematic / error variance
- Diff between variable, variat
- Principal axes
- What is principal in regression line? Fitting to data.
- Minimizing sum of square error

To do

Return to the paper we read first week of class, Monge
Look at the table and surrounding text
Ways to organize research hypotheses, analytic templates

Factor Analysis

Factor: a variat

We create it: a composite of all the measures on the other variables
Result of combining the other item measurements together
Not exactly a construct which is a label for a set of variables
Rather than a variable

Why use factor analysis?

Data reduction
Identify number of dimensions of a scale

Data reduction

Look more narrowly
You happen to have many items, data
You want to know what they all have in common
Getting rid of items that don't cluster

Identify dimensionalities

Sub-components that make it a fuller idea
Salience was hypothesized as 3 diff constructs
In the end, they determined that there were 2

Example constructs

Source credibility, variable in 60s,70s persuasion lit
- 3 dimensions: Trustworthy, knowledgeable, powerful/potent
Communicator apprehension

Exploratory factor analysis

Monge doesn't recommend highly
- Nevertheless, used often so one must be familiar with its use
Relatively flawed analytic device
Provides "infinity of solutions" rather than exact

Confirmatory factor analysis

All standard statistics available for determining significance
Exact solutions
Structured equation modeling
Not covered in this course

Factor matrix

Shows correlations among variables and factors

Procedures

Stage 1: Factor extraction

Principle component versus maximum likelihood
Standards for factor extraction
- Eigenvalue greater than 1
  - Eigenvalue: sum of correlation in a given factor
- Scree-plot: relative eigenvalue
- If value is > 1, we keep the factor
  - Scree test: Looks at the graph of a eigenvalues for steep slopes
    - Scree is a loose rock. When you step on scree, you slip
- Factors chosen and ordered according to the amount of variance
  - Up to ten factors, the tenth factor will be 100%

Stage 2: Factor rotation

Rotated v. un-rotated factor matrix
- Rotation should make interpretation easier
Orthoganal (independent) v. oblique (correlated) rotation
- Variat (orthoganal rotation)

Advantage of orthogonal solution:

Factors are independent of one another
Two distinct, unique, not correlated variables
Statistically independent

Time series, 24 Nov 2009

Issues in time series

Data always correlated
- Violating the assumption of most statistical tests we've seen so far
History is a cause
- Hopefully we have other variables that also account for the observed phenomena
Most over-time processes are nonlinear and must be treated that way
- General linear model: t-test, anova, regression
  - unit increase in x leads to general increase in y
- Instead, cyclic
- concommitant variation, multivariate time series
How do we determine causality when things vary together over time?

Different types of time series processes

Immediate
- Change in y occurs simultaneously
Gradual
- Change in y happens slowly
Delayed
- Change in y occurs later, requires good duration of time

Elements in a time series

Trend or slope (stationarity)
Cyclicality
Lag and lagged error or shock

Auto regressive integrated moving averages (ARIMA)

AR: autoregressive
- Degree to which variable is dependent on itself
I: integreted
- Degree to which there is trend
MA: moving averages
- Lag error or lag shock in the system

$Y t' = φ 1 Y t - 1 + φ 2 Y t - 2 + ... + θ 1 a t - 1 + ... - a t$

Simplest:

$Y t' = φ i Y t - 1 - a t$

Q-test

Forecasting

Residuals
Forecast horizon
Forecase confidence interval
Mean forecast error

In action

Collected data on 5 days

Built model on 4 days
Compared 5th day prediction with 5th day collected

Multi variate time series

Intervention analysis
- Experimental, giving an injection
- Seat belt law
Concomitant time series
- All data collected at once, hard to determine causality
Transfer function coefficients
Granger Causality
- Control for y history
Explained variance:

$R^{2}_{total} = R^{2}_{history} + R^{2}_{predictor}$