Posts on deblina

Making Mad-Libs

Fri, 21 May 2021 00:00:00 +0000

We played with an write-your-own-article / Mad-libs format a couple times of times at the The Dealer this year – see examples here and here – and it’s really nice illustration of basic jQuery functionality. This is a small tutorial of the code behind those articles. Major credits to R.E. for pioneering the form.

A run-of-the-mill input is pretty straight forward in HTML. (In the Dealer’s Wordpress backend, you can write HTML using the text tab on the classic editor.) For example:

Professor Richards' Words to Distinguish

Sun, 21 Mar 2021 00:00:00 +0000

In his syllabus for the Winter 2021 edition of the History and Philosophy of Science II (Late Renaissance through the Enlightenment), Professor Robert Richards made note of some commonly confused word pairs.

Refute vs Rebut

To refute is to actually prove something isn’t true
To rebut is to try and prove something isn’t true

Comprise vs compose

To comprise is to include or contain
To compose means to make up or form the basis of

The whole comprises the elements or parts, and the elements or parts compose the whole.

STAT 224 Notes

Sat, 05 Dec 2020 00:00:00 +0000

These are notes for STAT 224: Applied Regression Analysis, taught by James Dignam.

Lecture 3

Inference tests test the importance of the predictor

Lecture 4

Can do all the regular things (confidence intervals, etc) to $\beta$s
$R^2$ is a measure of explained variation

Lecture 5

MLR is regression with many predictors
Coefficients are one unit change holding all other predictors constant

Lecture 6

F-Test tests whether all the coefficients are simultaneously 0
Standardized residuals residuals have mean 0 and variance 1 and are dependent also
Internally studentized residuals are approx. normal and have mean 0 and variance 1

Lecture 7

Regression relies on assumptions about the model’s form and about the errors
- Mean of response is a linear function of predictors (linearity)
- Errors need to be normally distributed
- Errors need to have mean of 0
- Errors and resposne have constant variance over predictors
- Errors are independent of each other

Lecture 8

Ordinal and nominal categorical variables
- Represented in regression by dummy/indicator variables
Different categories need to be mutually exclusive
Omitted category is the base/control/reference category
Don’t keep interactions in the model without the corresponding main effects in the model (talk about effects of both, they go together now)

Lecture 9

Confounder is variable related to the predictor of interest and causally related to outcome but not in the causal pathway
Temporal ordering necessary in causal relationships
Effect modifiers are factors related to both predictors and response (they modify the strength of the association)
Control of confounding is key in explanatory modeling
- positive confounders overestimates the effect and vice versa

Lecture 10

When linear regression assumptions violated, tranform
- When transforming, all estimates and confidence intervals are expressed in that scale
Different violations call for different transformations
- non-normality is the least serious violation
- log transformation works when process/relationship is on a multiplicative scale
- square root form can have more stable variance over values
- Box-Cox Transformation is power transformation (find lambda)
  - Rank order preserving
  - $\lambda = -1$ is an inverse transform etc.

Lecture 11

Logit transform percentages ($\text{logit}(p)) = \log\frac{p}{1-p}$)
Reasons to transform:
- Unequal variance in the response variable
- Response variable not normally distributed (distribution the variable comes from has variance related to the mean)
Weighted Lease Squares (WLS) permits differential influence of data points
Autocorrelation is when error terms are correlated
- Still apparent randomness among the residuals, but some kind of serial pattern that suggests lack of independence (implies heteroscedasticity)
- Same observations in different time periods
- Spatially close observations
- Experiments run in batches
Durbin-Watson Test detects serial correlation
Cochrane-Orcutt transformation corrects for autocorrelated errors

Lecture 12

Multicollinearity is when some or all predictors in a model are correlated with each other
- Not an error; arises from the lack of independent information about predictor variables in the dataset
- Threatens interpretation that coefficient is the incrase in the mean of Y for one-unit increase in X
How much can we permit, and what do we do if there’s too much?
- All statistically independent means predictors are orthogonal
- If collinearity is strong and ignored, all CIs become larger and $\beta$s become unstable (change substantially when other variables are added/removed)
- Less of a concern for predictions; more of a concern for explaining a phenomenon
Signs of multicollinearity:
- F test is significant but all individual t-tests are nonsignificant
- A $\beta$ has opposite sign than expected
- General instability in estimates
- Large standard errors
Variance Inflation Factor (VIF), where J is a predictor $$\text{VIF} = \frac{1}{1-R_j^2}$$
- Greater than 10 is multicollinear by convention
What to do:
- Get more / better data
- Omit redundant variables
- Constrained regression
- Principal Components

Lecture 13

Variable selection:
- forward selection (add predictor with highest correlation with Y, then add highest partial correlation),
- backward selection (begin with all predictors, remove one with smallest t statistic), stepwise selection (add predictor as in forward selection, consider omitting based on backward),
- common approach (choose significant variables, put in MLR and remove non-significant, repeat until no more removing criteria)
Residual Mean Square is related to $R^2$ [\text{RMS} = \frac{\text{SSE}}{n - p}]
Use AIC and BIC to see if models are nested or not (those metrics balance information extracted from the data and number of parameters)

Lecture 14

Logistic regression is when response variable $Y$ is binary discrete variable (taking values 0 or 1)
- The model predicts the logit of the response as a linear function of predictors
Risk difference is the difference between one group and the other, as a proportion of the total
Odds ratio is ratio of odds of event in each exposure group
Difference of log odds is basic effect measure (same as log odds ratio)
Use MLE because error structure is different
- Null hypothesis is independence vs. association

Lecture 15

Even in logistic regression, each covariate is an adjusted effect meaning hold other predictors constant
Coefficients are log odds ($E^\beta_1$ is odds ratio)
Multiple logistic regression analysis:
- Likelihood Ratio – like F-Test for linear model, surfaces significant predictors
  - LR has $\chi^2$ distribution with DF = # of model parameters
  - As in SLR, redundant with test of only one predictor
- Hosmer-Lemeshow test measure goodness-of-fit (can id systemic variation that is not explained)
- AIC and BIC can also be used
- ROC curves plots sensitivity vs false positive rate (perfect prediction has area under curve of 1, best classifier is in upper left corner)
  - Can tell you what classification threshold to use (w. validation in independent samples)

Lecture 16

Generalized linear models are a framework for unifying theory and estimation models (need different: response, link function, error term, model)
- Link function addresses how linear predictor $\chi \beta$ relates to $E(Y)$
Poisson Regression (models count variables i.e. can’t be negative, as outcome)
- Assumption: conditional on the predictors the conditional mean and variance of the outcome are equal
  - Don’t use when variance exceeds the mean (overdispersed Poisson random variable) or too many zeros (zero-inflated Poisson)
- Use goodness-of-fit test to determine whether the model form fits data

Maverick Markets Notes

Mon, 20 Apr 2020 00:00:00 +0000

This is a guide to the double-starred readings that will be on the exam.

Hirschman, A., 1986, Rival Views of Market Society and Other Recent Essays. New York: Viking ch.5: Rival Views of Market Society, pp. 105-41

Covers different organizations of economy:
- Subsistence
- Fuedal
- Capitalism
Interpretation and evaluation of capitalism is historically contingent
- “The idea that the social order – intermediate between the fortituitous and the unchangeable – may be an important cause of human unhappiness became widespread only in the modern age” (105)
- Idea of changing social order to increase happiness is modern
- Coincides with idea that human actions can have unintended consequences
- Arguments rooted in existance
- Precedence of embededness
- Capitalism evocative (simple models implausible)
- Response based/dialectec transformation theory
The Doux-Commerce Thesis: commerce is a civiling agent of considerable power and range
- According to Hume and Smith, fosters industriousness and assiduity, frugality, punctuality, and probity)
- Market becomes cetral positon for the satisfaction of human wants
The Self-Destruction Thesis: capitalist society tends to undermine moral foundations on which society must rest
- Appeals to individual self-interest over collective interest (Capitalism has within it the seeds of its own destruction)
- Capitalism throttles all other institutions (commodifies everything)
Eclipse of the doux-commerce thesis
- Doux-commerce questioned and set aside in the latte rpart of the 18th century (industrialism wreaked havoc)
- Durkheim appears to generate a neo-doux commerce thesis, but is aware that any esnse of harmony generated by the division labor was an unintended consequence (superficial at best)
The Fuedal Shackles Thesis
- Contends that the capitalist revolution was never finished, left hanging as bourgeoisie market-bearers bowed to the demands of aristocrats and traditional authority figures
- Less than fully mature capitalist system (peripheral nations from World Systems theory – these theorists will try and explain why capitalism never matured)
- American exceptionalism: America came into being without the confines of a feudal past (also a curse according to Hartz – less ideological diversity)
This is a survey of different conceptions of capitalism – a coherant but complex analytic (tableau ideologique)

Rejects “over-socialized” view of some sociologists who think that choice is meaningless but also rejects the “under-socialized” view of economists who think that morals and values are not important
Creates concept of “embeddedness” – that economic action is “inside” social relations (i.e. larger structure provides opportunities and creates incentives)
- Uses the term “social networks” – and many attendant measures – to describe aspect of social structure that affects economic outcomes
Institution is durable way of doing things at national level (different from Stinchcombe, DiMaggio, and Powell, Scott, and Meyer and Rowan conceptions of institution)

Akerlof, G., 1982, “Labor Contracts as Partial Gift Exchange,” Quarterly Journal of Economics 97:543-69

Economics piece: notices that people work more than required, conceptualizes “gifts” to firm
- Gift in the sense of Mauss
Models gift giving in wage contracts
Conclusion: labor contracts are partial gift exchanges
- Wages determined by and influence worker’s effort and worker’s effort determined by norms (exchange)

Geertz, C., 1978. “The Bazaar Economy,” American Economic Review, May: 28-32

Focus on the role of information, communication, and knowledge in exchange processes (through bazaars)
- high degree of spatial localization and “ethnic” specialization
- search for information (not balancing options, but figuring out what options are)
Theorizes clientization (repeatative purchasers)
- Adversaries in the word for clientship relations that are not dependency relations
Notes paradox of bazaar buying: need to surround oneself with good communication links, but links are forged in antagonistic interactions driven by information imbalances

Biernacki, R., 1997, “Work and Culture in the Reception of Class Ideologies,“pp. 169-192 in J.R. Hall (ed.), Reworking Class. Ithaca: Cornell University Press.

Notion of culture in concrete research has misled investigators into embracing economically reductionist accounts
Offers alternative model of culture’s role in the crystallization of worker’s movements, a model that centers on the signifying practices of manufacture in the workplace
Cross-culural comparison of conveyance of power/commodification in labor (labor power in German context, embodied labor in British contexts)
Analytic he proposes “theorizes history [but also] historicizes theory”

Taxonomy of the social life of money: gift (a person’s voluntary bestowal on another), entitlement (a right to a share), and compensation (direct exchange)
- each corresponds to a significantly different set of social earmarking techniques to distinguish those categories of social relations (earmarking and control are key!)
- People sort ostensibly homogenous legal tender into distinct categories
Bureaucratization, Commercialization, and monetization significantly altered the scope, form, and content that taxonomy of money
- Studied discretionary payments in the sexual economy, discretionary payments in bureaucracies and routinized payments in bureaucracies

Baker, Tom, 2001. “Blood Money, New Money, and the Moral Economy of Tort Law in Action.” Law and Society Review 35(2): 275-319

Qualitative study of personal injury lawyers in CT, implications of professional norms and practices that govern tort settlement behavior
- Moral and practical barriers to collecting “blood money” (money from indvidual defendents, as opposed to liability insurance companies which give “insurance money”)
  - Very little changes hands, but important social tool
- Challenges conventional understanding that tort law in action is simpler, more streamlined than textbook tort law
  - Client is liability insurance company
  - Taking blood money requires more justification than a simple mistake (moral and practical explanations)
Compensation and retribution figure far more prominently in tort law in action in action than does the deterrence emphasized in much of the theorhetical and doctrinal literature

Helleiner, E., 2003, “One Market, One People: The Euro and Political Identities,” pp.183-202 in P. Crowley (ed.), Before and Beyond the Euro. London: Routledge.

Examines Euro’s role in eroding national identities and developing pan-European identity (one money to one market, but one people?)
According to Simmel and Marx, money turns traditional social and personal ties, into ones characteriseed by impersonal and instrumental economic calculations
Some key ideas:
- Imagery on money and the construction of political identities
- Currencies as a medium of social communication
- Collective monetary experiences and communities of shared fate
- Money and popular sovereignty
- Modern money as the fiduciary standard

Velthuis, O., 2003, “Symbolic meanings of prices. Constructing the value of contemporary art in Amsterdam and New York galleries,” Theory and Society 32:181-215

Arguing for a multiplicity of meanings in prices (not just signaling/profit as in econ, not just contaminating or corrosive as in humanities)

CS 152 Notes

Sat, 14 Mar 2020 00:00:00 +0000

These are notes for the final exam in CS 152, taught by Diana Franklin. The code for the class can be found here.

Data Representation

To convert to unsigned binary:

Number of places is number of bits, fill all with zeroes by default (every place has value of 2 to the power of the place – from the right, zero based indexing)
Find highest power of two that goes into decimal number
Go to that power’s place, and change to one
Keep adding powers of two until you get to number

To convert to hex:

Introduction to Science Studies Notes

Mon, 09 Dec 2019 00:00:00 +0000

The 16 readings here were the required ones featured on the exam.

Merton, Robert K. 1942. “The Normative Structure of Science,” pp. 267-278

Early text in science studies — analyzing science as an institution, trying to identify norms.
Merton is an institutionalist (institutions in the sense of trust and behavior) — so studied science as a collection of internal rules and scripts (“norms”)
- Matthew effect
- Norms can be studied by sociologist

There are four main norms that constitute science: universalism, communism, disinterestedness, organized skepticism. (CUDOS)