Podcast
Questions and Answers
Within the framework of Jack Vroomen's ethical perspectives, which statement most accurately differentiates between 'what IS ethics' and 'what ISN'T ethics'?
Within the framework of Jack Vroomen's ethical perspectives, which statement most accurately differentiates between 'what IS ethics' and 'what ISN'T ethics'?
- Ethics assesses immediate consequences, while non-ethics evaluates long-term implications.
- Ethics involves an impartial viewpoint considering the general public's welfare, whereas non-ethics stems from an individual's subjective perspective. (correct)
- Ethics is concerned with identifying unacceptable behavior, whereas non-ethics only considers what is desirable.
- Ethics focuses on personal moral judgements, while non-ethics addresses broader societal implications.
Given the principles of consequentialism, which of the following scenarios presents the most nuanced challenge in its application, particularly when assessing whether consequences are 'good for everyone'?
Given the principles of consequentialism, which of the following scenarios presents the most nuanced challenge in its application, particularly when assessing whether consequences are 'good for everyone'?
- A government implements a policy that leads to economic growth but increases income inequality. (correct)
- A community decides to preserve historical landmarks, which restricts new housing developments.
- A company adopts sustainable practices that reduce profits but enhance environmental conservation.
- A hospital prioritizes emergency care, leading to longer wait times for non-urgent patients.
How does Kant's categorical imperative, specifically the formulation that one should 'Act only according to that rule whereby, at the same time, you believe it should become a universal law,' pose a critical challenge to practical ethical decision-making in complex scenarios?
How does Kant's categorical imperative, specifically the formulation that one should 'Act only according to that rule whereby, at the same time, you believe it should become a universal law,' pose a critical challenge to practical ethical decision-making in complex scenarios?
- It struggles to accommodate situations where universal application leads to paradoxical or undesirable results. (correct)
- It relies on subjective interpretations, undermining its objectivity.
- It necessitates predicting all possible outcomes of an action, which is often infeasible.
- It assumes a universally shared understanding of what constitutes a 'good' outcome.
In what way does virtue ethics shift the focus of ethical evaluation, and what critical challenge arises from this shift, considering the perspectives on rules in deontology and consequentialism?
In what way does virtue ethics shift the focus of ethical evaluation, and what critical challenge arises from this shift, considering the perspectives on rules in deontology and consequentialism?
Considering the distinction between 'Part of ethics' and 'Part of methodology' in research ethics, how would you categorize addressing potential conflicts of interest that might bias the interpretation of research results, and why?
Considering the distinction between 'Part of ethics' and 'Part of methodology' in research ethics, how would you categorize addressing potential conflicts of interest that might bias the interpretation of research results, and why?
Within Karl Popper's normative methodology, what fundamental critique underlies the caution against scientists pursuing self-interests that might compromise the falsifiability of their work?
Within Karl Popper's normative methodology, what fundamental critique underlies the caution against scientists pursuing self-interests that might compromise the falsifiability of their work?
How does Benford's Law, which describes the non-uniform distribution of first digits in naturally occurring numbers, serve as a tool for detecting data fabrication, and what critical assumption underlies its applicability?
How does Benford's Law, which describes the non-uniform distribution of first digits in naturally occurring numbers, serve as a tool for detecting data fabrication, and what critical assumption underlies its applicability?
In the context of detecting influential observations, what distinguishes DFFits (Difference in Fits) from DFbeta (Difference in Beta) in regression diagnostics, and how does each contribute to assessing model stability?
In the context of detecting influential observations, what distinguishes DFFits (Difference in Fits) from DFbeta (Difference in Beta) in regression diagnostics, and how does each contribute to assessing model stability?
When addressing multicollinearity in regression analysis, what are the trade-offs between using Principal Components Analysis (PCA) to create new variables and employing Ridge Regression, especially concerning variable interpretation and bias reduction?
When addressing multicollinearity in regression analysis, what are the trade-offs between using Principal Components Analysis (PCA) to create new variables and employing Ridge Regression, especially concerning variable interpretation and bias reduction?
How would Shapley values be applied to interpret the relevance of individual variables in a multiple regression model where several predictors are highly correlated, and what specific advantage does this approach offer over traditional coefficient-based interpretations?
How would Shapley values be applied to interpret the relevance of individual variables in a multiple regression model where several predictors are highly correlated, and what specific advantage does this approach offer over traditional coefficient-based interpretations?
In the context of missing data, what critical distinction differentiates 'Missing Completely at Random' (MCAR) from 'Missing at Random' (MAR), and how does this distinction influence the choice of appropriate imputation methods to mitigate bias?
In the context of missing data, what critical distinction differentiates 'Missing Completely at Random' (MCAR) from 'Missing at Random' (MAR), and how does this distinction influence the choice of appropriate imputation methods to mitigate bias?
How does the presence of autocorrelation, induced by interpolation of missing data points in a time series, fundamentally challenge the validity of ordinary least squares (OLS) regression, and what are the implications for statistical inference?
How does the presence of autocorrelation, induced by interpolation of missing data points in a time series, fundamentally challenge the validity of ordinary least squares (OLS) regression, and what are the implications for statistical inference?
What are the implications of temporal aggregation when transitioning from high-frequency to low-frequency data, particularly concerning the validity of statistical inference and the interpretation of model parameters, and how can the geometric lag model address these considerations?
What are the implications of temporal aggregation when transitioning from high-frequency to low-frequency data, particularly concerning the validity of statistical inference and the interpretation of model parameters, and how can the geometric lag model address these considerations?
Under what conditions should researchers be most concerned about the presence of spurious relations in regression models, particularly when analyzing time series data, and what strategies can be employed to mitigate the risk of drawing incorrect causal inferences?
Under what conditions should researchers be most concerned about the presence of spurious relations in regression models, particularly when analyzing time series data, and what strategies can be employed to mitigate the risk of drawing incorrect causal inferences?
Within the context of 'blinded by the data,' how do additive outliers (AOs) critically compromise the parameter estimates and model specification in autoregressive conditional heteroskedasticity (ARCH) models, and how can researchers preemptively address these issues?
Within the context of 'blinded by the data,' how do additive outliers (AOs) critically compromise the parameter estimates and model specification in autoregressive conditional heteroskedasticity (ARCH) models, and how can researchers preemptively address these issues?
What are the primary sources and implications of the 'Optimism Bias' and 'Planning Fallacy' in forecasting the completion time of infrastructure projects, and how do these phenomena collectively undermine accurate cost-benefit analyses?
What are the primary sources and implications of the 'Optimism Bias' and 'Planning Fallacy' in forecasting the completion time of infrastructure projects, and how do these phenomena collectively undermine accurate cost-benefit analyses?
In forecast adjustments, how do the 'Law of Small Numbers' and 'Optimism' pose cognitive challenges for experts, and what systematic biases do they introduce when subjectively refining model-based predictions?
In forecast adjustments, how do the 'Law of Small Numbers' and 'Optimism' pose cognitive challenges for experts, and what systematic biases do they introduce when subjectively refining model-based predictions?
In the context of Customer Relationship Management (CRM), how might insurance companies leverage customer behavior data, and what ethical considerations arise when using predictive models that may inadvertently perpetuate discriminatory practices?
In the context of Customer Relationship Management (CRM), how might insurance companies leverage customer behavior data, and what ethical considerations arise when using predictive models that may inadvertently perpetuate discriminatory practices?
How might reliance on Artificial Intelligence and machine learning raise ethical concerns, and why are the ethical Al guidelines necessary?
How might reliance on Artificial Intelligence and machine learning raise ethical concerns, and why are the ethical Al guidelines necessary?
Why are 'active inclusion' (incorporating diverse data) and 'fairness' (ensuring equitable outcomes) critical in the development and deployment of AI systems, particularly in contexts where algorithms may inadvertently perpetuate or amplify societal biases?
Why are 'active inclusion' (incorporating diverse data) and 'fairness' (ensuring equitable outcomes) critical in the development and deployment of AI systems, particularly in contexts where algorithms may inadvertently perpetuate or amplify societal biases?
Flashcards
Ethics (Philosophy)
Ethics (Philosophy)
Systematic reflection of morals, underlying principles to determine what is ethical.
What IS Ethics?
What IS Ethics?
Considering what is good for the general public from an impartial perspective.
What ISN'T Ethics?
What ISN'T Ethics?
Looking at ethics from your own point of view.
Moral Point of View
Moral Point of View
Signup and view all the flashcards
Moral Judgements
Moral Judgements
Signup and view all the flashcards
Consequentialism
Consequentialism
Signup and view all the flashcards
Utilitarianism
Utilitarianism
Signup and view all the flashcards
Deontology
Deontology
Signup and view all the flashcards
Kant's Categorical Imperative
Kant's Categorical Imperative
Signup and view all the flashcards
Main Difference 1/2
Main Difference 1/2
Signup and view all the flashcards
Ethics Is Not Only Negative
Ethics Is Not Only Negative
Signup and view all the flashcards
Virtue Ethics
Virtue Ethics
Signup and view all the flashcards
Basic Idea
Basic Idea
Signup and view all the flashcards
Research Ethics
Research Ethics
Signup and view all the flashcards
Part of Ethics
Part of Ethics
Signup and view all the flashcards
Part of Methodology
Part of Methodology
Signup and view all the flashcards
Normative Methodology
Normative Methodology
Signup and view all the flashcards
Criticism
Criticism
Signup and view all the flashcards
Post-Popperian Methodology
Post-Popperian Methodology
Signup and view all the flashcards
Where do the data come from? and How do they look?
Where do the data come from? and How do they look?
Signup and view all the flashcards
Study Notes
- Jack Vroomen defines ethics as the systematic reflection of morals
- Ethics involves underlying principles to determine what is and isn't ethical.
- People have different moral intuitions based on socio-economic background.
- Ethics involves looking from an impartial viewpoint, what is good for the general public
- What isn't ethics involves looking from your own point of view
Morals
- Aim for impartiality
- Moral judgements require looking at the whole spectrum: what is wrong/bad or acceptable/desirable
- Moral values include fairness, justice, and equality
Reflection
- Involves formulation of general ethical principles
- Sometimes reflection affects your own moral compass
- Other times, it criticizes it
Ethical principles
- Consequentialism: actions should be good for everyone
- Utilitarianism: actions ideally maximize collective utility, where everyone is worth the same (Bentham/Mill)
- In utilitarianism there arises the issue of assuming each other's welfare can be compared
Deontology
- Actions ought to be done out of duty with respect to others because their rights should be respected
- Kant's categorical imperative: act only according to the rule whereby you believe it should become a universal law, do not treat others solely to serve your own ends
- Issue with deontology is that it lacks the possibility for exceptions, and the question of whether the law is a good one if everyone follows it.
Main difference between consequentialism and deontology
- In consequentialism, only consequences count, and not the motives
- In deontology, only motives count, not the effect
- Both generally lead to the same actions as they reflect a concern for others
Ethics
- Not only negative
- It also considers what kind of person one wants to be, and what kind of society one wants to live in
- Virtue ethics focuses on people and not on actions
Virtue ethics
- People with good character traits are considered ethical
- A issue with virtue ethics is what virtues, and how do they interact with each other
- Proponents of virtue ethics usually make their own list of virtues
Basic idea
- If you do not keep on exercising your attained virtues then you lose them
- Both consequentialism and deontology focus on rules, which can always be evaded/bent/abused
- Virtue ethics focuses more on the root, which is difficult to change
- The cons of virtue ethics is the are too vague and imprecise
Research ethics
- "Acceptable/desirable behavior in scientific research" where desirable is stronger than acceptable
- Research ethics focuses more on methodology
Research ethics and philosophy
- Research ethics is at most partly part of philosophy ethics
- Not an issue: are consequences of research practices for the interest of others
- Is an issue: when is research conducted in proper way?
Two parts of research ethics
- Part of Ethics: direct scientific malpractice can have a societal impact, indirect malpractice can impair trustworthiness of science, impairing democracy in the long run
- Part of Methodology: aims to distinguish malpractice/misconduct from proper scientific work and exists prior to societal impact.
Clear malpractice
- Plagiarism
- Fabrication/distortion of data
- Neglect of ("inconvenient") data
- Impairs trustworthiness of science, slows down scientific progress
Normative methodology
- "Proper/best methods to use and standards to invoke in judging whether some research practice is truly scientific?"
Normative?
- A part of the philosophy of science different from the ethics
- Not to describe but to prescribe and make it the norm
Karl Popper
- Falsification: clear evidence should refute a hypothesis with the example that all swans are white
- Not exempt from confirmation bias
- Scientists want self interest to make academic career
- Scientists do not want papers to get refuted because that makes you a bad academic
Post-Popperian methodology
- Thomas Kuhn and Imre Lakatos use peer reviews
- Science as self-corrective enterprise, only if right institutions in place
- Criticism: a big career is not made for performing peer reviews
Keywords
- Advice, people do not typically understand what you are doing
- Advisor, we are the advisors of policy makers, as econometricians
- Choices, should be clear what choice you make, report everything
- Consequences, every choice has consequences
Lecture 1
- With good research practice, if you search for some result, you can find it, which is not the proper was to conduct research
- There are 14 general ethical guidelines, most important is collegiality (allowing to share data), good reference practice, and availability of results
- There are also ethical guidelines for statistical practice
- Suggestive graphs can be used to provide various conclusions
- P values are the probability of observing an extremer outcome/result than we have witnessed in the sample, given that the null hypothesis is deemed true.
- Note: if a null is rejected, do not try to interpret the p-value. all that can be said is that if i reject the null, i must conduct more research
- P hacking: misuse of data analysis to find patterns in data that can be presented as statistically significant by performing many statistical tests and only report on the ones that come back as significant.
- P hacking is when i have a result that looks promising, i stop and skip the rest of the data
- Harking: Hypothesizing After Results Are Known: it is not good to base the initial thought on the results of the paper. rather, include these thoughts in the discussion.
Size and power
- Type 1 error (alpha): reject H0 if H0 is true (preferably small, never zero) (also known as size)
- Type 2 error (beta): not reject the H0 if HO is not true (preferably small)
- Power of a test (1 - beta): reject H0 if HO is not true (preferably large)
- A DNA database is an example of multiple testing where even if the test is accurate you will find false positives as long as the sample size is large enough
- Correlation is a mutual relationship between variables
- Causality is a mutual relationship between variables where one causes the other, and seek for thing that incorporates correlate to whether the results disappear
Lecture 2: Fake data
- It is easy to generate data that gives you a fit you want Diederik Stapel calls, "create study after generating data → harking”
- According to Benford's law Humans have difficulties creating random numbers, resulting in the Frequency of the first digit of a number is which decreases as digits become larger
- Very large t values mean results are too good to be true
- Very small p values are the same as t values, but using a different methodology
Lecture 3: Influential observations
- Never delete data prior to analysis
- An additive outlier occurs in case of dickyfuller and large values tests, which makes any evidence of a unit root disappear
- The delta will be biased to zero, removing the unit root if the large spike in is not well predictable by yt-1, and again makes it a bad predictor
- it is bad to want a result based on a single influential observation
Detection
- Detection of influential observations involves deleting each observation separately and looking how model changes
- Anscombes quartet are 4 totally different datasets that have the EXACT same regression line when using OLS, which means it's important to always take a look at data visually before running diagnostics
Outlier diagnostics:
- Rstudent: dummy for each observation separately and look at t-statistic of dummy
- Diagonal elements of Hat matrix X(X'X)invX' to look for leverage of observations: expect to be around the mean leverage which is k / n (k = #regressors; n = #observations)
- DFFits; difference of fits (r-squared) if observations removed. >| 2sqrt(k/n)| is significant at 5% level
- DFbeta; difference in beta of an observation (when removed), with it being significant if > |2/sqrt(n)|
- Random walk is unit root in time series data
error correction model
- Add the stationary linear combination of the cointegration related time series to correct for this error
- Recursive residuals run a model with which has residuals, where incremental data creates an extra observation, and the residual is reported again
- A sudden change in the recursive residuals indicates an influential observation
- Recursive parameters are the same but for beta and test statistics are reported
- See outliers at forecast of origin, see if it creates a new level, is a type, or an innovation
- Winsorizing is when any observation above/below a certain quantile forced to this quantile observation.
- Trimming discards the extreme values
Lecture 4: Model selection
- It is not good to assume there is true DGP, so instead do model averaging
- Running tests on different model can increase error unless using bonferroni correction
Model selection
- Diagnositc tests like AIC and BIC
- General to specific: start with many variables and move/remove variables that increase AIC/BIC the least
- Specific to general: start with a small model and then add the variable that decrease AIC/BIC the most
- Prefer general to specific because it prevents omitted variables:
- Problem: G2S and S2G may differ in final model outcomes, use cross-validation!
Stepwise methods
- Split test data and run cross validation to prevent over fitting outliers
- Cointegration
Lecture 5: Estimation and interpretation
- Running a counterfactual exercise with the example of 9/11, we want to compute the loss made, which is the data from before 9/11 to make forecasts and compare those to the actual values during 9/11
- Seasonal unit roots can better be s separate time series
Growth curves/s-shaped data
- Logistic
- Gompertz
- Bass
- combining forecasts results in better forecasts if models are not nested
- Forecast accuracy is improved, removes possible interpretation
Combining the forecasts
- Averaging
- weighted average (based on quality of model insample)
- Bayesian
- The lower the BIC the higher the weight
Measurement error:
- measurement error is unobserved
- The FIX is to use Instrumental variables
Instrumental variables
- Using an exogenous variable to separate the endogenous part of a variable away from the exogenous part, and then run the regression with only the exogenous part
- The exogenous variable must have no correlation with the error term with explanatory power that resembles x to some degree
Two stage least squares
- Regress endogenous variable on exogenous variable, and get estimates of endog var as gammahat’z.
- Regress your normal regression with the estimated version of the endog var.
- Simultaneity is when two variables have influence on each other at the same time, which causes endogeneity
- A Variance inflation factor test variable correlation
Multicollinearity
- Makes the variables look statistically insignificant while they are significant due to inflated standard errors; due to regressor being dependent on other regressors
- Only a real problem if the multicollinearity is very large for the forecasting
Fixes for multicollinearity
- Delete one of the multicollinear variables
- ridge regression
- PCA, summarizes the variables to create new variables
- Quadratic regression includes demeaning the parameters
- Principal Components Analysis
Shapley values
- Used to interpret a number from a model and determines the relevance of the variables to the model
- Accomplished by running regressions with all combinations of explanatory variables and where the shapely value is weighted average of all R^2 for regressions that include a certain regressor
Lecture 6: Missing data
- Attrition happens especially in panel data, where people do not want to participate in a prolonged study.
- OR will start to behave differently: Not all data is collected for them or they start to behave differently because they feel like they are being watched.
- Missing completely at random is not problematic if the missing mechanism is independent from observable vars and estimates
- Fixes are interpolation or set to average of the set with low variance
Types of Missing Data
- Missing at random: missing mechanism is dependent on one or more of the observable variables - Younger people are less likely to survey questions
- Missing not at random is a big problem and ex: smokers may not fill in health aspects of smoking. The fix is to ask better questions or using random surveys.
- Adapt estimation procedure (so ask a similar question and then estimate) or interpolate
- Interpolation causes autocorrelation to be non-zero, where few interpolated points result in arge auto correlation
- Auto correlation measures dependence with lag varaibles
- Persistence is measuring how long the auto correlation lag has on following observations
- Randomized rotating panel: use multiple panel groups that you rotate over time, prevents attrition and gives less bias
- Selective sampling is a method to estimate parameters when you have a choice model with a lot of zero's in the data
- Aggregation is modeling consumer behavior
- Transition matrices are not observed
aggregation
- Temporal aggregation is when data is high frequency aggregation
- Is often-considered a vehicle to measuring current and effect
- Koyck model
- MIDAS model
- Mixed data sampling is averaging of the data with no auto correlation
Lecture 7: Spurious relations
- Spurious relations are the statistical relatsionships that are not actually there
- It is a “mathematical relationship where two or more events or variables are associated but not causally related "with coincidence
- Include trends in regression models to get rid of spurious relations and check Auto correlation
- Due to the large amount of data a structural brea is easy to obtain
Lecture 8: blinded by the data
- Suggest a model with a cosine function
- Check data and make prediction intervals
- Additive outliers (AO): AO is enough to make ARCH misspecify the parameters
- ARCH autoregressive conditional heteroskedasticity model
- Get of AO before doing arch
- Take care of potentially misspecified conditional mean
- Structural breaks can give non lineartu
- If you don't account for the structural break, then the model can suddenly give a trend a big parameter
- Robust estimation models allow for structure breaks in the modelling
Lecture 9: Predictability
- Infrastructure projects like the Olympic Games have contractors that do not put the actual high cost due to completion.
- Planning fallacy: optimism bias in estimating the future cost.
- Select sample data that is still relevant
- Reconciliation of groups = Forecast
- Trends are known
- Structural break can mean the change trends
- One must specify the trend of the data to model
- Limits to predictability is when the model reaches the general means of data
Lecture 10: Adjustment of forecasts
- Model-based forecasts when properly done, expert adjusted forecasts can be more accurate
- Expert adjustment if an expert knows to interpret a model, with their knowledge to adjust the Forecast
Law of Small Numbers
- in a casino, people see that the number of times that "red" was landed on is hugh and then their expecation of "red" next goes up
Forecaster behavior
- Optimistic: people like forecast things will goes well
- Growth curves
Forcast Updates
- the KLM case, forecast updates were used to smoothly model
Lecture 11
- Customer Relationship Management is using data
- High level transformations
- Clustering happens base don distance
- Zip code aggregations
Lecture 12: Machine Learning
- Hierarchical Bayes method should get the data to copy
- Artificial Neural Network
- Discrimation uses data in the algorithm
- ethical Al guidelines:
- Include ethical Ai
- Fairness to conclusions
- Algorithem must be replicatable
Studying That Suits You
Use AI to generate personalized quizzes and flashcards to suit your learning preferences.
Related Documents
Description
Ethics is the systematic reflection of morals, involving underlying principles to determine what is ethical. It aims for impartiality, considering fairness, justice, and equality. Ethical principles like consequentialism and utilitarianism guide actions to benefit the general public, maximizing collective utility.
