6. Prediction & Causal Inference - challenge

[JunsolKim]

Post your response to our challenge questions.

First, pose a causal research question you would like to answer (in one, artfully worded sentence ending with a question mark). This could be the same question you posed for any prior week's assignment, or a new one that improves on or updates it, or shifts it to the counterfactual context. Second, identify a counterfactual prediction that will enable you to make your inference. This counterfactual prediction could simply be the question itself (e.g., How will the stock price change if the CEO reveals/admits an overstatement of earnings?) or it could support or validate the answer of that question (e.g., How do I predict whether the sentiment of a given sentence is positive or negative, certain or uncertain, resonant with U.S. Republicans or Democrats, about environmental position X or Y, etc. IF some prior event Z occurs). Finally, describe the datasets on which you will perform your counterfactual causal inference. Parenthetically note whether this data could be made available to class this week for evaluation (not required...but if you offer it, you might get some free work done!) Please do NOT spend time/space explaining the precise model or analytical strategy you will use to generate, evaluate and utilize your prediction. (Then upvote the 5 most interesting, relevant and challenging challenge responses from others).

[konratp]

Causal research question:

• With the right-wing extremist AfD entering the German parliament in 2017 has led to a shift towards more conservative buzzwords being used in parliamentary speeches.

Counterfactual:

• In other German-speaking countries like Austria and Switzerland, right-wing extremist parties like the FPÖ (Austria) and SVP (Switzerland) have been represented in parliament for a much longer time. If there, too, the political tone has shifted right-wards, that might be evidence that the developments in Germany are not because of the AfD entering parliament.

Dataset:

For the analysis of this question, I would be drawing on data from the open discourse project, accessible here.

[GabeNicholson]

How does Covid related news cause changes in government policy toward Covid? The counterfactual trend could be finding large policy changes and how they relate to the timing of exogenous news articles. Perhaps states that previously were reluctant towards restrictions suddenly became interested in them after certain article topics became prevalent.

The dataset would be the Covid data set in the Large Corpus Collection.

[Jasmine97Huang]

Causal Question: How does lyric affect the popularity of a song? Counterfactual: If lyrics have not impact on the popularity of the song, it might indicate linguistic blindspots for music consumption and hence limit the significance of lyric as units for cultural analysis. Data: Same as previous weeks

[pranathiiyer]

Does the reach of a newspaper affect the kinds of castes, education levels, and income levels people mention in matrimonial ads in newspapers? counterfactual: If there isn't a difference in the ads published by people from the same states in newspapers with varying reach, it could mean that this isn't really a function of the reach/distribution. (the obvious caveat is here is that it might not be the exact same set of people publishing in various newspapers which might have to be dealt with more qualitative analysis).

Data: same as previous weeks, ads from the tribune newspaper in india

[ValAlvernUChic]

Do mentions of transient workers in newspapers affect the intensity of online xenophobic behaviour?

Counterfactual: If xenophobic behaviour on social media remains constant, even if fluctuating mentions of transient workers, positive or negative, then there could be something else more important.

Data: Newspaper data from NOW corpus available but not the corresponding social media mentions

[Qiuyu-Li]

The causal question: Does prevailing sexism lead to a high gender wage gap? The counterfactual prediction: An otherwise identical county with less sexist people has a lower gender wage gap. The dataset: I can use the frequency of tweets labeled as sexist in a county to proxy the “sexist level” of this county, but I didn’t come up with how to use text to construct the confounding variables. A more direct way is to just use traditional data, like local GDP, demographics, etc.

[Sirius2713]

Causal question: Does the company-naming tweets of Trump cause the stock price fluctuations of the corresponding companies? Counterfactual: How did the stock price of the corresponding companiy change when Trump mentioned a company on Twitter? Dataset: Same as last week.

[facundosuenzo]

• Causal research question: The often-called "scandal" between Facebook and Cambridge Analytica impacted online newspapers' coverage of technology issues to shift their framing from "benevolent" to "evil."

• Counterfactual: If the scandal would not impact the news coverage, we should see 1) a similar framing of the two-time frames 2) explore possible confounders (like the news source and its political relation with Facebook).

• Dataset: The dataset I'm working with is NOW.

[mikepackard415]

Causal question:

• Are the topics discussed in environmental discourse affected by the news of the day?

Counterfactual:

• If a major news story relating to food and agriculture becomes popular, it fill be followed by several in-depth analyses on the same topic.

Dataset:

• The same data I've been working with, the corpus of articles from environmental magazines. Can be made available!

[NaiyuJ]

Question: Do preferential policies about non-political issues (such as education, economics...) make ethnic minorities in China happier? Counterfactual: How ethnic people's attitudes and emotions change when adopting new preferential policies about non-political issues or just discussing certain preferential policies. Data: Baidu Tieba posts by ethnic minorities in China

[LuZhang0128]

The question is: In an online social movement, activists can spark conversations between multiple audiences by linking discussion topics usually discussed in isolation. The counterfactual is: Cross-topic posts only have weak influences since most online social movements' topics are so well-developed. Audiences all have their primary focuses. Activists could be better off if they only focus on their primary interests. Data: Twitter data containing hashtag #BLM.

[hshi420]

Question: Does the number of fake news increase when world events or crisis happen?

Counterfactual: The world events and crisis offers opportunity to the fake news makers to create conflicts between countries.

Dataset: Can combine a fake news dataset and a timestamp that shows when there was a world event or crisis.

[Emily-fyeh]

Causal Question: In Taiwan, supporters of the losing party show a weaker national identity after the election. Their social media posts tend to deprecate their nationality and their democracy. Counterfactual: If the election result would not affect the sense of national belonging, then the patriotism sentiment within the social media posts would not fluctuate.

Data: Facebook/Twitter data

[kelseywu99]

causal question: Do digital news outlets help with making fake news accessible to the public? counterfactual: If fake news stories appear, then it has a high probability to appear in a digital-only newspaper instead of a brick-and-mortar press. data: the fake news corpus I mentioned for the previous week

[chentian418]

Causal question: Does seeing value-relevant news of overall positive sentiment induce sell-side analysts to revise the earnings forecast upward? Counterfactual: When there is a negative signal in the instant news about a firm, the corresponding analysts will revise their earnings forecast downward.

Data: Proquest news data and analyst and firm-level data from I/B/E/S

[chuqingzhao]

Causal Question: Do visionary pitches or pivots leads to early-stage future success? Counterfactual: if firms do not make visionary pitches or pivot innovative ideas, venture capitalists may underestimate the firms' potential and make conservative investment decision. Data: Crunchbase data

[sudhamshow]

Question: Does extensive content moderation (flagging and blocking) and switch to another that doesn't? (Thereby creating echo chambers) Counterfactual: Noticing that people stay back and make their voices heard even louder and rally for attention when one of their prominent influencer/group member is banned, instead of shifting base to another platform.

Data: Reddit community content moderation and admin activity data

[YileC928]

Causal question: Does the sentiment of investors' social connections affect their own perception of an asset? Counterfactual: If the connections of an investor are bearish, he tends to adjust his expectations downwards for his own portfolio. Dataset: same as past weeks.

[Hongkai040]

Causal: Does the gender of commenter influence the number of upvotes they receive for their movie comments? Counterfactual: How does the number of upvotes of movie comments vary when the gender of commenters are different? Dataset: same as last week.

In fact, we can also replace 'movie comments' with 'reading responses' ... So the dataset will be all reading responses and upvote records in this repo or in the Computational Social Science Workshop repo...

[sizhenf]

Causal: Does the issue of critiques affect their chance of being censored? Counterfactual: What is the difference in censorship rate among critiques on different issues, controlling for their emotions?

Dataset: same as last weeks

[isaduan]

Causal: Does nationalist discourse in science policy affect actual policy outcomes? Counterfactual: The policy discourse proceeds policy changes

Dataset: same as past weeks

[ttsujikawa]

Question : I'm interested in seeing how cultural background affects our ways to build relationships with people. I would focus on distinct seasons of the Japanese reality shows, terrace house to see how people behave differently in a culturally diverse setting. Counterfactual: Does cultural diverse setting affect people in a way that people become more extrovert and active for building relationship?

Data: Netflix

[ZacharyHinds]

Question: Does the use of Incel-specific slang/identity-narratives in forum posts affect user engagement? Counterfactual: Incel posts with more slang/identity-narratives have the more commenter responses.

Dataset: Archived post/comment data from Incel forum, same as past weeks