Stats Notes 2

2012-02-07T11:16:00.000-08:00

Reliability Problem
--two counters next to one another detect different things.
--One counter picks up the same thing over and over.

Accuracy--How close to true value
Precision--How reliable your measurements are

Citing does not prove reliability or credibility.

National Land Cover Database--has a lot of error associated with it, and is based on a sample--actual data points are geo-located, so you can sum by county.

Reliability--all your instruments measure the same thing. Google Street View and pedestrian counts. "Four passes on the same street at non-standard times, on non-standard days"

Inter-relator reliability--the people who related it to you are consistent between two different people.
Expert panel of 10, with a common definition of each, then had them review scenes.

HLM -- nested within one another to reduce correlation.

Descriptive vs. Inferential Statistics.
Descriptive--"not very interesting".
Income used as a variable to be explained, or to explain another variable.

Average income.
Income in relation to income.
Income, controlling for gender, and relation to income.

Gender as a MODERATOR.
Education a causal variable in explaining income.
Causal variable = determinant

Public health has rigorous protocols so as to establish data consistency--to ensure the data collected is comparable.

Confounder: Variable that is correlated with determinant, causally related to dependent variable.

"On the causal path between the dependent and independent variable"

INTRODUCTION
METHODS
RESULTS
ANALYSIS
CONCLUSION
-At end of every article, talk about the limitations of your research. "It appears that" not "People generally".

"If you want to get a good deal, you have to be willing to walk away".

VALIDITY
-Face, Construct, Internal, External

"the way he operationalized (measured) the construct of smart growth--the same proportions of urban, suburban, and rural"
--if you go to the portland definition of smart growth, you are actually becoming more sprawling.
"Who says Portland is Smart Growth?"
-A Face Validity issue.

1 guy, surveying towns in Idaho--reliability issue, only 1 inter-relator.
-Tried to explain it in terms of income.
-Operationalized 'Civic mindedness' as church attendance and voting.
--We know civic minded people who don't go to church, and church-goers that aren't very civic minded.
Face validity: Would ten people in a room agree that this is a good measure.

Construct Validity:
Sprawl: Fairly high uniform density and lack of functional open-space.
"Not valid measures of the construct of sprawl".

Internal validity - One things causes another.
External validity - ability to generalize the results.
--can i use Portland to explain Atlanta?
--Lots of wonderful data for NYC--applicable to SLC?

Smaller open spaces--different from large ones.
Inferential Statistics--you know nothing about the population, but can infer (from a sample) something about the population, with a certain level of confidence.
Confidence Levels and Confidence Intervals
95% certain that it falls in range of 2-5.

"5% chance that I fail to reject the null hypothesis"
Significance level-chance you got the result by chance.

Four Types of Variables:
NOIR
Nominal, Ordinal, Interval, Ratio
Nominal=name
Ordinal=Rankable (Likert Scale)
Interval=Number that can be added/subtracted.
Ratio=Number that can be divided. Has a true (absolute) zero. Income.

Operationalize constructs with variables--different variables measure only parts of a given construct.
Equivalency Reliability
If we use both, we will have an issue of multi-colinearity.

"If I had independent measures, I might use factor analysis (principle component) to create a new variable....The common variance of our 7 variables.. what is common to all of them".

"I don't understand this in a statistical way, but I do understand how to apply them" -Reid.

Unexplained variation = error term.
Face validity--it just seems right

Mediator on Causal path
Correlation is not R^2... is R, so need to square it to get the percent explained.

Must have a true zero to calculate an elasticity.

Use age categories as dummy variables, because of non-linearities.

f statistic is t statistic squared?

Stats Notes 1

2012-02-07T08:38:00.001-08:00

Question everything--root of research.
-Counterpoints of induced travel and sprawl is good.
Roadbuilding & VMT -- which is cause and which is effect?

'The preponderance of evidence suggests'...but some suggests the opposite.

Accessibilit..

Check it all empirically.
Theories of travel, but cannot check a conceptual model. Use data to check the conceptual model.
"When you write your dissertation, you will have to have a conceptual framework".

Most things in the world, which we operationalize--travel, capacity, growth--abstract concepts.
Operationalize with variables: VMT, highway lane miles, growth w/ pop_growth, sprawl w/ residential density.

Variables
Dependent--predicted
independant - predictors
explanatory and explained
x and y

Theory
-explanation of why thigns occuur
-relates constructs to one another
--"if highway capacity increases, peoiple will travel more"

Hypothesis
-Statement about relationships derived from theory
-theory cannot be directly tested
-"as highway lane miles increase, VMT will increase"

Law
-Theory confirmed by many tests of related hypothesis
-"Universal law of traffic congestion"
-Tony Downs (old planning), his law not in fact a real 'law'.

Population
-set of people/things/places with characteristics you wish to measure

Sample
-subset of population
-assumed to be selected at random
-inferential statistics relate samples to population

DATA
-Values of variables
-Aggregate of disaggregate data
-Aggregate--done using averages.
-Disaggregate--by traveler or household.
--Viewed as much more reliable.

Data-sets
-Cross-sectional - At this point in time. Same question accross a population?
-Longitudinal - changes over time
-Mixed - cross sectional and longitudinal
-Panel - same people, over time.

Scales: Nominal, Ordinal, Interval, Ratio
Nominal (Categorical)
Ordinal (Rankable) - Best, Middle Income,
Interval (Continuous) - GPA, Celseus
Ratio (True Zero) - Income, Kelvin

Obese "Categorical" - Binary, over or under a given thresh-hold.

Unit of Analysis
-What is being analyzied
--Individuals, groups, social organizations.

Unit of Observation
-Entity for which you collect data.
-Census: HH
-The census tract
-May aggregate up from a lower level of observation.

Function--relationship between dependant and independant variables

Correlation: Index of STR of relationship
Causality: When one thing actually causes another.
-Can only be established with controlled experiments,
-Required to establish: Statistical association, time-order, and elimination of rival variations.

We do actually do true experiments-Portland congestion pricing experiment.

Only going to meet 8 times over the semester.

"Controlled experiments seldom done in planning because it requires assigning people at random to control groups".
-Non-random assignment confounds this.

-Stated preference of stated choice surveys.
'Experiment' where you posit something--"If we build a new rail line--20 min @ $4, or 20min @ $2 min.
-Must be done as forced choice.
_If you just ask "Would you use it?" people say 'sure'.
-Forced choice in preference to stated preference.

Don Appleyard - actual experiment, peopled driving by at different speeds.
Likert Scale - "On a 1-5"

Model - anything that represents anything
Mathematical Model -
-uSe mathematical language
-functional relationships between variables

Simulation Models - STELLA
Statistical Models - Correlation Relationships

Elasticty- 100% increase in variable #1leads to x% increase in variable #2

Ceteris Paribus - "All else being equal".
-Self Selection bias - Avenues vs. Draper.

Will read Reid's paper on elasticities, and then Jason Cow's. @ U of Minnesota--he will create data set for us.
-His paper talks about controlling for self-selection.

Phillip -
Shima- Column on self-selection, and Jason cow's paper.

"Toward a better understanding of influences on physical activity"
Bauman, Robert Wood Johnson Foundation, San Diego State
-Amir to read and present.
--Paper talks about different types of variable.

Planners care about causality--conceptual plausibility required.

Invariant Law -- Law of supply and demand is not--some exception cases. Law of gravity is invariant law--no exceptions.

In sprawled area - more pollution, but lower exposure. In compact, people are living in areas of high-impact.

Read Reid's columns--in our research methods book, converting one into another.

Correlates-Variable correlated with another
Determinants-Determines the value of anotehr
Mediators - Intermediante between determinants and outcomes
Moderators - Variables that moderate or affect the relationship between determinants and outcomes.
Confounding -What we percieve as the determinant is actually a correlate.

Sprawl-->Inactivity-->Obesity
Determinant-->Mediator-->Outcome

Overview of Quasi-Experimental Research

Reid will send me that column. --will send original and revised.
-Present the results of the ___________& Lewis paper. 5 minutes, couple of slides.

To control for:
1) Random assignment
2) Quasi-experimental design
3) Statistical study (focus of this class)

Good Noise

Stats Notes 2

Stats Notes 1