10/15: Andrew Stier and Marc Berman

[shevajia]

Comment below with questions or thoughts about the reading for this week's workshop.

Please make your comments by Wednesday 11:59 PM, and upvote at least five of your peers' comments on Thursday prior to the workshop. You need to use 'thumbs-up' for your reactions to count towards 'top comments,' but you can use other emojis on top of the thumbs up.

[skanthan95]

Thanks for sharing this work, looking forward to your presentation! I was wondering whether for the study, you had knowledge on whether the individuals in your sample(s) had mental health conditions that are often co-morbid with depression (like anxiety), and whether they were taking any medications (i.e., antidepressants) -- given that these factors could act as potential confounds. I'm also curious about whether you might replicate this study in a covid-context, to see how the unique extenuating circumstances of the pandemic might impact the relationship between city size and depression prevalence.

[lulululugagaga]

This is really an interesting topic and the studies for depression is no doubt urgently-needed. Do you plan to do other psychological studies related to urban and rural topics? It seems that suicide is also an important issue and it's kinda related.

[bjcliang-uchi]

As a social network analysis fan myself, I am really looking forward to this presentation! My question about this research, and perhaps many research papers that use similar methods, is: how to think about the difference between a simulation (based on probability and degree distribution, such that nodes are considered essentially the same) and a null model?

[hesongrun]

Thanks for presentation! This is a very interesting research. I am wondering if you have investigated the mechanism through which large cities may associate with less depression rate? Are there any insights that can be provided to city planning in improve the overall mental health of the dwellers? Thanks!

[xzmerry]

This research is interesting! My question is about data usage and the possibility to generalize the research.

1, In the paper, it mentions that "It suggests that larger city environments and urbanization may actually provide greater social stimulation and connections that may buffer against depression", but there could be selection bias for social media data, so some people (such as younger people or who have more free time) may use twitter more frequently than others, and thus causing systematic bias to the research, how do you cope with it? 2, Besides, how to ensure that big data used here is as accurate as data of traditional surveys? How to validate it? 3, Moreover, I feel like the research is mainly for people in the city and under urbanization, but could this also incorporate twitter users in the countryside? Cuz I think this is an unsolved problem for both big data and traditional survey methods, considering people in the countryside(especially rural area) are often under-sampled. Thanks

[YanjieZhou]

Thanks very much for your presentation. In my opinion, the pressure for competing and the abundance of entertainment and social connections in cities act as two opposite parties in the term of depression. Apparently, here positive parts take the edge. But I think we cannot just control socioeconomic conditions and reach conclusions, as socioeconomic statuses are very important in impacting our depression. So could you please elaborate more on different impact of social connections according to people's socioeconomic statuses?

[MegicLF]

Thank you for discussing such an interesting topic. I am curious about how you consider the impact of the latent relationship between two persons, which is unobservable but would create noise in the measurement of depression. Also, several nodes in the network may form a connected graph, then there may exist an accumulative impact between all people inside this connected network. How would you estimate this issue?

[xxicheng]

I have a quick one: do you have any plan to expand your research to a wider range? Like different countries. How are you going to deal with the cultural difference in this way? Or do you think that different cultures will influence the results? Thanks.

[ttsujikawa]

First off, I got amazed by how you have built up your hypothesis and tested it using a creative way to examine the association between urbanization and mental issues. Both are certainly growing phenomena, so it is super interesting to dive in. Here is a quick thought of mine on this research. The paper states a little bit about the influence of COVID-19, but I think its impact on how we work and live is going to be enormous in a sense that people may start shifting their work style into remote work. This global trend may allow people to leave urban areas and live wherever they want. At this time, it is crucial to think about how this deurbanization or digitalization of work affects people's mental health. I would love to hear your opinion on this. Thank you!

[shenyc16]

Thank you for sharing your research with us. I find the result that number of depression cases has negative correlation with city size very interesting and it is totally opposite to my previous thoughts. As you have mentioned in the paper, you conducted population level method and examined scaling relationships of mental health from the perspective of cities as networks. As a student exposed to this idea for the first time, I am wondering how to comprehend and rationale this general analysis, like how can this analysis work after ignoring the intricate and personal details? How to account for other important factors, i.e. GDP growth rate or crime rate, based on the empirical model in the paper?

[Qiuyu-Li]

Thank you so much for such interesting research, and I'm very impressed by how psychology can be combined with big data techniques! Just one quick question: In my understanding, you attempt to verify the positive correlation between city size and depression level. However, I fail to find evidence that this correlation results from the socioeconomic network. Would you be able to tell us more about that? Thank you!

[Lynx-jr]

Thanks for this inspiring paper!

I mainly have 3 questions about this paper, one is the same as @Anqi-Zhou -- how to quantify/estimate the emotion in the tweets?

The other 2 question comes from an observation from daily lives - @william-zhu-3541 mentioned, cities have better entertainment and cultural life. If a city has consistently expand its administrative borders throughout the years, and the outskirts of this city, although highly developed in terms of infrastructure (has newer and larger facilities), have a less dense population than the city center area. Also, people living in the suburbs were counted into the "city" population but do not have sufficient access to "better services and amenities" or "better entertainment and cultural life". Hence, they are expected to behave differently and more likely to be depressed than those who live in the city center. This resonates with the conventional opinion that depression is associated, at the individual level, with fewer social contacts. So the second question is how to effectively reduce such internal effects.

Lastly, I agree with what @siruizhou said, economic status and income inequality should be taken into consideration when examining mental health issues and this somehow reduces the effects I just said. How can these factors be implemented in this model?

[mintaow]

Really enjoy going through how you approach the psychology topics from the perspective of spatial characteristics!

I also learn a lot from the method to manipulate geographical data. I am curious why you decided to roll up county-level data to Metropolitan Statistical Areas (MSAs) instead of other proxies such as Commuting Zone (CZ) developed by Tolbert and Sizer (1996). I have heard there is some paper (e.g. Adrien Matray (2013)) are preferring using CZ rather than MSAs, since the geographical boundaries are not constant over time in MSAs.

Thank you!

[WMhYang]

Thanks for the paper. To be honest I am not quite convinced. I think the relationship is reversed, i.e. people with more connections are those who have the ability to connect with more people. Therefore, it might be the case that they are more outgoing and optimistic so that they can have more connections. I apologize if I miss something in the paper, but could you please give me some hints on how to alleviate my concerns? Thanks again.

[zixu12]

Thanks for sharing your research! The peers of mine have raised many great questions such as @siruizhou and I cannot wait to hear your opinions on that. A question from me is: as the context of this paper is the United states which is a well developed country that has experienced urbanization and counter-urbanization. What about counties that is still in the developing stage, or the countries that has experienced urbanization but not counter-urbanization yet? Will your conclusions still hold for these countries or what can be referenced from your conclusions? Thanks!

[sabinahartnett]

Thank you for sharing your research Andrew and Marc! As this research is studying people in a snapshot in time (/the past when collecting twitter data ... and digital footprints can often be edited), I would be curious about following individuals as they move between urban and rural areas. Did you consider how urban and rural lifestyles may be conducive to different ways of using social media and communicating? As someone new to a city, I thought about how transitions and how long an individual has been in a location may also have an effect on your data. Looking forward to the presentation!

[Rui-echo-Pan]

Thanks for sharing. It's interesting to see this counter-intuitive result as revealed about the larger city with lower depression rates. Apart from that, I would be more curious about the mechanisms in it. I guess there are complex mechanisms influencing the lower depression rates, some of which may be positive or negative. Would you say it's interesting and applicable to explore, say among the similar size cities, (probable large cities), what heterogeneous factors would lead to different depression rates?

[YijingZhang-98]

Thanks for sharing this excellent research. I noticed that the data sources you used for the depression prevalence estimate come from two different agencies, with the different collection and reporting methods. For example, as you mentioned, "NSDUH is conducted in person while the BRFSS is conducted over the phone", the participant with severe depressive symptoms may refuse the in-person survey, which results in a selected sample comparing the other one. Could you provide more details on how you deal with the two data sources that make it comparable? Thanks!

[Yaweili19]

Thank you for your presentation in advance! Looking forward to learning from it. I especially like your approach to model depression by experience and social network, which I think is very innovative yet makes great sense. I would like to learn more about the theoretical background of this model and what are the inspirations you draw from to create these interesting models.

[Panyw97]

Thanks for sharing! I am curious about your analysis

Users with lower numbers of tweets are more likely to have depressive sentiment in their tweets

How did you come to this conclusion? Will it be affected by the numbers of tweets, but not only by the sentiment effects?

[ChivLiu]

Thank you for the presentation and the wonderful research! Following @harryxiong13 's question, from Alonso's Bid-Rent theory, people living between the urban areas and the suburban areas are usually those who suffer the most from maintaining their living conditions. I wonder if we think about the urban areas in large cities such as New York and Chicago, people living very close to the CBD are usually young, single, and ambitious, and people living in luxury suburban usually have achieved something and have a stable family. Therefore, is there any evidence that people from specific age groups would be more possible to have depression?

[fyzh-git]

Thanks for bringing this intriguing topic. I have a few questions about two details.

1. (p4) 'This expectation is directly observed in cell phone networks and indirectly via the faster spread of infectious diseases such as COVID-19, and by higher per capita economic productivity and rates of innovation.' How can higher per capita economic productivity and rates of innovation measure the connections and socioeconomic interactions?
2. (p4) 'We will now need to pay particular attention not only to the average number of social connections in a city of size N , k(N), but also to its variance across individuals in that city and how they influence depression.' Do we really need to look into the individual details in order to generalize the model? As mentioned in (p10), 'the fact that important insights about the mechanisms of mental health disorders might be gleaned from such a general population level analysis, which ignores the intricate and often personal details of mental health, is surprising and powerful.' Do they conflict with each other? Thank you!

[RuoyunTan]

This is really interesting. I am actually one of those who indeed prefer larger cities, mainly because I feel that in big cities there are more outlets for my negative feelings. So I can totally resonate with the part where you said, "larger cities provide a buffer against depression".

But on the other hand, I wonder if there are any innate features of large cities that are unaccounted for but can affect your model. For example, I would guess that larger cities have a higher level of mobility. In small cities, you are measuring the level of depression of a relatively stable group. But in cities like New York and Chicago, people come and go and this may affect the measurement results.

[vinsonyz]

Thank you for your presentation! I really enjoy this topic. I think it would be interesting if we have the data that people move from urban to suburban and from suburban to urban. How would you justify the causal effect in your research?

[jinfei1125]

Hi, my question is a little similar to Jade's one because we recently have been learning the properties of observational data in Perspectives class 😄 In the article, you mentioned you explored a large geo-located Twitter dataset of individuals and their messages for depressive symptoms in different cities over one week in 2010. So I wonder how you make your data is representative of all the population in NYC? What if some positive people are more likely to speak out on Twitter?

[luxin-tian]

Thank you for sharing. I wonder how would computational psychologist address the selection bias in the measurement of depression rate. Since the diagnosis is conditional on the patients' actively seeking for medical attention, there can be selection bias if the intention or consciousness of reaching to medical care is heterogeneous across geographical regions.

[weijiexu-charlie]

Thanks for your presentation. You found lower depression rates in big cities, suggesting that large cities may provide a buffer against depression. However, I'm still curious about what's the underlying mechanism that leads to this negative correlation. That being said, what makes big cities have a lower depression rate? and how? I also believe that the pandemic must be able to serve as a natural experiment that has applied some sort of treatments on our society to help us test some potential hypotheses. For example, the pandemic may cut off the connection at the neighborhood level to some extent and might have eliminated the contrast of city and rural life in some aspects. How do you think that we could take advantage of the pandemic to understand the mechanism behind urbanization and depression?

[Jasmine97Huang]

Super interesting work and finding! Although I agree with the many "skepticism" about information published on social networks, using Twitter data supplementarily to the two traditional surveys data is still a great attempt to reduce reporting bias.

However, I am questioning if Twitter is the best social platform for observing depressions. Why not using Google Search data, which might help avoid social desirability bias? Additionally, instead of using machine-learning to detect messages for depressive symptoms, a sentiment analysis may be more appropriate to evaluate and emotions behind these messages.

[luckycindyyx]

Thank you for sharing such an interesting paper that use economic methods to tell a psycological story. I had a similar question with @ydeng117. In many metropolises (such as New York, Paris and Beijing), people live in the suburbs (or even in another city) at night, and work in the city center during the day. So the distribution of population in a city would vary largely with time in a day. Is that possible if any time factor or distribution parameter being included in the model as well to make the problem more precise? Thanks!

[goldengua]

Thanks for your presentation, I am interested in the way you detect depression from Twitter data. Language posted on Twitter can be an unfaithful account of one's emotional status and very likely to be dependent on the habit of users and audidence. On the other hand, depression can be a complex state to quantify with a binary variable. While I can totally understand getting access to medical records is extremely difficult, I was wondering how you do content analysis to address the complexity of depression classification problem.

[lyl010]

Thank you for sharing such an interesting research with us! It is inspiring and I love the part of generative social networks and the association between individual social network degree and depression probability. So I am looking forward to learn more details about the generative social networks. About the model and the interpretations of results, I have a few questions:

1. How to decide the parameters in K(N)? How is its robustness across different datasets?
2. I like the bright side you tell us about the large cities, since we already knew too many dark sides, as those you mentioned. I am wondering if there is any way to interpret this theoretical contradiction?
3. As @Nak Won Rim said, the number of depression cases is associated with node's average degree and the scale of network, and the depression is a binary variable. But when depression is a distribution, the polarization of mental health might be more server as the scale of social network increases(since the distribution of degree might be more 'power law'). Would this change the present results?

[adarshmathew]

So this paper caught my attention for its focus on mental health and the use of networks, two things I care deeply about.

Two questions, to add on to the excellent ones raised by @bakerwho, @wanitchayap, and the brilliant @nwrim:

1. The following line from page 4:

.. the risk (probability) for an individual to manifest depression, taken to be inversely proportional to their social connectivity,

Isn't this an extremely stylistic assumption in your model, given your generative network style? Especially since you haven't indicated why that should be the null/base case. In cities driven by the gig economy, I may 'interact' with multiple people, but that may not be a meaningful, personal interaction that has significant bearing on my mental health. How would you justify this when it comes to generalizing your result?

2. This ties to your use of the Twitter dataset and overlaying the PHQ-9 questionnaire on it, and a larger question about dataset/framework re-use: As @JadeBenson pointed out, there are multiple contexts of use for social media content, often vastly differing from the sober context in which patients answer PHQ-9 questions. How do you justify using this dataset + framework to estimate user-level mental health indices? Projects like UoVermont's Hedonometer attempt to track moods at a large, macro scale using Tweets. But in your context, geo-tagged tweets can tell you, at best, what a neighborhood is feeling on average if there is a collective spike/fall at the same time. Are tweets really suitable for the use of the PHQ-9 methodology?

[yongfeilu]

Thank you very much for sharing! I am very curious about how you discern the emotion behind tweets on twitter considering that users can post very subtle words that might have the opposite meaning to the expression. Moreover, could you explain more how you choose the specific models and techniques to do the natural language process job? Thanks!

[robertorg]

Thanks in advance for your presentation. My question would be if you think large cities in developing countries would also exhibit lower depression rates compared to smaller ones. Also, I was wondering if other factors usually associated with big cities, such as better medical resources for mental health, would be more important than socioeconomic networks.

[k-partha]

Interesting data and methods. Is there similar data exploring the relationship between happiness/life satisfaction and living in cities? It would be interesting to compare the results - perhaps they might suggest a more complicated picture. Another question that immediately pops is whether depressed persons choose to remain in/move to in rural areas. The average psychographic traits of an urbanite are likely to differ from that of a farmer. Do you think that accounting for these differences might add significant value to this study?

[romanticmonkey]

Thank you for your presentation. Through your studies, what would you suggest of urban as well as rural mental health workers to improve on the current status quo?

[aolajide]

It would be interesting to use the "framework for conceptualizing and modeling mental health in complex physical and social networks" to analyze the oft-publicized and oft-researched correlation between poor mental health and digital/social media use.

Aditya Shukla. (2019, August 14). The Effect Of Social Media On Mental Health And Well-being. Cognition Today; Cognition Today. https://cognitiontoday.com/2019/08/effect-of-social-media-on-mental-health-well-being/

‌

[JuneZzj]

Thank you for presenting in advance. I am impressed by the point of "high population density with a higher incidence of depression has been mediated by a higher density of negative social connection." Do you think people they actively choose to live in that region tend to have different behavior or mental health than those who choose do not live in those areas? Thanks.

[luyingjiang]

Thank you for your presentation. It is quite interesting and popular nowadays. I am wondering how you consider the differences between the major metropolis and small cities? Will the results be affected if we switch the location (i.e. to a different country)?

[cytwill]

Thank you @nwrim for this interesting paper. I love the part where social networks and psychologic theories combined together to generate a computational social science project. However, I have some questions about your establishment of the model.

First, there are many strong assumptions such as some mentioned by @adarshmathew. Though I understand this is quite common in many simulation-based social network studies, it still proposes some doubts about why it is necessary to do so. If they are necessary to generate the hypothesis, why don't you propose the hypothesis directly but bothering to use this network?

Second, as shown in the research, your findings disagree with some of the previous findings. Have you thought about any intermediate variables or mechanism that can probably explain such differences? Or, if these differences can be related to different data sources of your choices?

Thanks!

[YaoYao121]

Thanks for bringing us this wonderful sharing session.I'm curious about your research vision. How you found this interesting topic and how you decided on involving computational methodologies in it.

[XinSu6]

Very insightful work and I found the connection between the urban science and psychology very fascinating to read about. I have work in sentimental analysis before in a internship but it was far less attractive than what this paper have discusses. My question here would be how are you quantitively distinguish and measure the the emotions here and how do you know that the it is accurate to do like this.

[chentian418]

I am impressed by your improvements to avoid biases in reporting the number of depression cases, which not only include two annual population surveys but also implicit cases based on passive observation deploying corpus analysis on Twitter datasets.

Meanwhile, I am also curious that whether the modeling of the number of depression cases is accurately enough to catch the exact relationship. on the one hand, city size alone as a main independent variable may be not convincing enough to explain for variations in number of depression cases, and I would think of the average number of social interactions a good variable to interact with the models, either a direct variable in the model or a channel through which the city size affect the number of depression cases. on the other hand, the exponential model is indeed easy to interpret in economic language, but how do you justify this relationship?

Moreover, I was thinking that the per capita prevalence of depression in U.S. cities may be a too broad population to study and may undermine the causal relationship we built. For example, there might be cases that people feel stressed out in large cites and move to small ones for rehabilitation, while they still remains depressed when surveyed in smaller cities. Scenarios like this made the study for per capita prevalence of depression not a so meaning index, and we may consider study the influence of city sizes on some specific populations that have less noise or more meaningful to study.

Thanks!

[mingtao-gao]

Thank you in advance for your presentation! The paper presents an interesting finding. My question is related to the external validity of the experiment, will the result be consistent if similar data collected in a different cultural seting is applied to the model?

[timqzhang]

Thank you for your paper ! I have similar question as before, that is the potential bias in the tweet spidering of emotions, since emotions are hard to judge. Does this way have some convincing algorithms?

[caibengbu]

Thank you for sharing such an interesting paper that use economic methods to tell a psychological story. I do think per capita prevalence of depression in U.S. cities may be a too broad population to study and may undermine the causal relationship we built. I would love to hear more about your explanation during the presentation and see if I missed anything.

[Yiqing-Zh]

Thank you for sharing the work. I am wondering whether there could be some factors that simultaneously affect people's choice of living in different regions, their social network, and their depression level so that there might be some endogeneity problem when we refer to a causal relationship. Thank you!

[Tanzi11]

I am looking forward to your presentation, and I wonder from a policy standpoint what cities can do to further ensure the mental health of their communities based on your findings. It would be interesting to also see how the digital divide between generations affects people in cities differently.

[YileC928]

Thanks for bringing us this wonderful research! There are two questions that I am curious to learn about:

1. Would rural-urban mobility influence your findings?
2. What kind of applied implication (e.g., for public policy) do you think this research can offer?

[ziwnchen]

Thanks for sharing this interesting article! It is really inspiring to see that, contrary to popular belief, urbanization help decrease the depression rate! I saw your research use the generative social network to explore how social exposure influence mental health--how the real-world social network could be represented by the generative model?