Open HyunkuKwon opened 4 years ago
Would the measurement for structural embeddedness be able to account for positions like admins, HR, etc? Employees in these positions, by nature, often send out many emails to many people within the company. Hence, they would, by the method chosen, have a very high structural constraint. However, I feel like these positions actually have a low structural constraint; they interact and contact many people in many departments (but with often weak relationships) so they have the information advantage. Would it be better if they exclude these positions? Or it would be better to choose a different measurement for structural constraints?
Metric for structural constraint: I agree with @wanitchayap in that the discussion of the metric for structural constraint deserves more than one sentence (that's literally how long the discussion is). In particular, I'm interested in how the metric can adjust for the direction of the exchange. Building onto Mint's argument, an HR manager or administrative staff may be frequently "spamming" other employees, but they most likely do not receive a response for most of the emails sent. In that case, is it justifiable that they are assigned a high level of structural constraint, as predicted by Mint?
Furthermore, I wonder why it takes second-degree exchanges as an input. I calculated based on the information of "10 mil emails of 600 employees over 5 years ..." and concluded that on average an employee sends or receives 300 emails a week. So I am guessing the inclusion of second-degree exchanges is not motivated by concerns about sample size? But if one happens to interact with a highly structurally constrained individual, she would also be assigned a high degree of structural constraint, is that a desirable outcome?
Metric for cultural fit: I am not sure if I am fully convinced with this, either. Oof. In the discussion section, the authors divulge the reasons why they hesitate to declare any causal effect, one of which is also the reason why they insist to make within-firm comparisons, that different firms may have very different "cultures". I believe the same argument can be applied to different departments or positions within a given firm. For instance, there may be a culture in the R&D department that encourages prioritizing innovation, which may reflect in their emails in the form of topical or linguistic (or even emotional, don't have a strong prior for this) difference. In the case, an individual with low culturally fit may in fact be culturally fit, but this is not captured since being culturally unfit according to the metric is what makes one culturally fit in reality. This is an unnecessarily long example but I just have much doubt about these two metrics, which are at the core of the paper. Personally I am most interested in how cultural fit changes across time: Does it increase steadily? Does it vary with structural constraint as one understands the trade-off through experience? Unfortunately, time only serves as a fixed effect for hazard rate, and used to allow individual characteristics to vary over time, if I understand correctly.
It seems that the employee's English writing level is not accounted for (such as a native speaker or not). The cultural embeddedness is measured by the differences of stylistic, topical, and emotional characteristics between outgoing and incoming emails. Thus, employees from various regions might inherently have different embeddedness.
Another interesting perspective would be looking at how employees' embeddedness evolve as they have more email interactions with others: whether employees who got promoted also "learned to fit" the culture better?
This is an interesting paper which investigates the tradeoffs between structural and cultural embeddedness. I have the following two questions:
(1) cultural fit will promote (inhibit) attainment for individuals with low (high) network constraint, and (2) network constraint will promote (inhibit) attainment for individuals with low (high) cultural fit.
(Continued) However, I think the greater-than-one coefficient before the intersection in Table 3 only provides evidence for the first fold because the coefficient before network constraint is also larger than one and cultural fit is a measure greater than zero. I would draw a conclusion that network constraint will always inhibit attainment for individuals, regardless of cultural fit. Where am I wrong?
The second question regarding the difference between incoming and outgoing emails. It is commmon that the reply of an email is much shorter than the email itself. For example, a manager may just give a one-sentence comment to an employee who has sent a long essay considering the recent development of something. Hence, Those who receive emails a lot and who send emails a lot might have inhabitant difference in the measure of divergence. However, I do not think this should count as difference in cultural fit. How could we take this into account?
This paper is a quite good approach to dig out a kind of dilemma between fitting in and standing out. It also poses the question to us to think that under such a tradeoff problem, how individual make their decisions.
My first question also considers the metrics of culture fit. It is with no doubt that the size of the email corpus is sufficient for analysis with about 10.24 million email messages, but I would like to mention that the emails in the firm are usually different from normal emails that could extract some culture patterns. Emails in the firm are mostly professional emails, usually without much discussion on private issues or emotions, but with much rules of wording. For example, it is likely that the firm has developed a kind of implicit message principle when writing the email to colleagues, and we may not find much divergence from that. From this perspective, what are some methods to alleviate this pattern which may cause bias? Maybe an exploration of whether such a wording "formula" on the corpus exists could precede the main-body research.
This question follows the first one, which also echoes the second question from @WMhYang. There may exist two types of unbalance among the emails. The first one is what WMhYang mentioned, the unbalance of the length of email between sender and receiver. Moreover, it is also worth noticing that the emailing between employees is not linear as simple as A-B-C...etc, but a complex network among all the employees. Given this scenario, there are so many conditions needed to control before measuring the cultural fit. For example, different people may have their major email senders who send them emails most frequently, or each people have different weight on the emails sent by various departments. It creates another kind of unbalance. Therefore, is it still justified to measure the cultural fit by the divergence of incoming-out emails in general, as it may be hard to make good comparison based on different conditions?
This is really a thought-provoking paper that provides me with creative approaches to measuring structural and cultural embeddedness. My questions are as follows.
In the paper, the cultural fit is measured by the divergence in the semantic categories in an individual's outgoing and incoming messages. As mentioned by @WMhYang and @timqzhang, I wonder whether such an approach would be confounded by the natural difference (such as the length of a message) between incoming and outgoing emails?
The authors employed involuntary exit as the dependent variable due to the availability of data. From my understanding, "involuntary exit" stands for being expelled and it is intuitively difficult (at least for me) to distinguish between the effects of the structural and cultural embeddedness. Would it be better if we instead track whether an employee gets promoted in the organization?
The methods mentioned in this paper to extract structural and cultural embeddedness information in a huge amount of emails are really impressing, which give me a lot of inspirations on my final project. My question is similar to the second one of @WMhYang. I think the tone people use when they speak to different people or respond to different issues will be different. In different periods, when people have different connections to other people (for example, in the application season students may have more emails with the admission committee but in the summer they may spend more time with their friends), which may cause the difference of their reaction in the emails. How do we consider this change?
@liu431 raises a great point - as Professor Keysar from UChicago psychology department's work puts it, using a second language can "blunting emotional reactions associated with the violation of deontological rules." (or at least make them communicate in a different style). Also, even without considering English as a second language, I think the degree in which people can communicate with their first language varies across people - not everybody gets full marks on SAT, GRE, etc. I feel like this difference in fluency could affect metrics that depend on texts (like the cultural fit metric in this article) - is there any way we could account for this issue (in metrics or in NLP in general?)
Another thing that I am curious about is that while self-report measures get thrashed a lot, their validity is actually quite good (but I am mostly thinking in a psychology perspective now, I admit). I think to test the metric with other self-report measures of cultural fit to see if they correlate with one another could potentially be good support for the validity of the metric. Is this something that is done regarding text-based metrics?
On metrics for structural embeddedness: this paper used Burt's constraint, the measurement for structural holes, which is meant to measure "the extent to which time and energy is concentrated within a single cluster". In a firm, departments are clustered groups by nature, and there will be positions like HR, tech services that are meant to build bridges across groups, but their low constraint index do not represent their uniqueness in a firm. The paper used fixed effect model controlling for department afflications, but that would not resolve the concern 100%, as it is more about position, than department. Can we evolve the model to a directed graph to alleviate this problem? For example, HR might receive a lot of request emails, whereas assimilated brokers might be initiating most of the paths. (It would be great if authors could present more descriptive figures about the network...)
The measurement of network constraints and cultural fits are very interesting. This paper contends that the cultural fits and network fits are exactly one thing: balancing between social belongings with differentiation(part: conclusion). I really appreciate this effort since it helps me to better understand the idea of structural constrain: the position of broker is in fact the result of effort of differentiation. And brokerage is in fact a good thing: people who not embedded constrained in a structure have lower probability of involuntary exit than those who embeds deeply in their structure.
My question is why people who have constrains from both structural network and cultural network even face higher probability of exiting involuntarily than those who do not embed in either network? Disobedience is in fact a good thing, and is it a good result for manager?
The paper assumes that an actor's cultural assimilation is captured by the difference in sentiments / linguistics in her outgoing messages vs her incoming messages. The authors note that this measures "the extent to which an actor is culturally assimilated with her active interaction partners, whom she is primarily dependent on for productivity and who are most likely to evaluate her performance".
I'm not sure I agree with this assertion. In many firms, those who provide performance ratings are often not the people an employee might interact most with, or whom they necessarily need to be "culturally assimilated" with. In a consulting firm for example, an analyst (entry level employee) might interact and get directions from senior analysts, but be evaluated by principals / partners because "management" has to provide the ultimate review of someone's performance. There's a natural separation in the hierarchy of an organization, and the principal/partner may not give weight to how culturally assimilated the analyst is in her team. How might this confound this analysis?
As many of my colleagues pointed out about the accuracy of operationalizations of two kinds of embeddedness, I'm interested in asking about the operationalization of attainment. I think involuntary exit is one extreme case of negative attainment that happens relatively rarely. I'm wondering if there is any other way of measuring that?
This is such an interesting read. My question is related to the second conclusion, that is 'network constraint is associated with higher attainment for individuals with low cultural fit'. How should we interpret this conclusion in real life? Does this mean that people who don't talk like others in their groups are more likely to be more important group members or leaders, who therefore have higher attainments?
Does the level of attainment differ by department? Would someone in technology be promoted faster than someone in sales for example, and be retained for longer as well? I am wondering if there are distinct groups that might impact the attainment levels.
Also, what about emails sent outside of work? Personal emails between colleagues? These might have a lot more to do with cultural fit than work related emails.
This paper provides an interesting insight in statistical relationship between cultural fit and network alignment. I am curious about whether word choices can perfectly capture the cultural perspectives embedding in it. I am also wondering about the extrapolation of the findings. How much could it extrapolate to other findings?
The method and models used in this paper are innovative and insightful. As many students mentioned above, to what extent can the variables, both dependent and independent, faithfully reflect structural embeddedness and cultural assimilation remain a little unclear. Another point I am highlighting is the company's organizational culture environment, especially when we are applying the theory in this article into today's high-tech companies. It is widely acknowledged that the top popular tech companies in labor markets have self-distinguished cultural features and these are sometimes playing a critical role in employees' decisions. In a word, though the database in this article seems to be abundant, it is better to be compared with datasets from other similar companies.
The Goldberg et al. paper demonstrates how one might analyze social interactions using network theory and the centrality measures discussed by Borgatti. Several unique aspects of this study stand out and present further questions:
My question is why people who have constrains from both structural network and cultural network even face higher probability of exiting involuntarily than those who do not embed in either network? Disobedience is in fact a good thing, and is it a good result for manager?
Post questions about the following orienting reading:
Goldberg, Amir, Sameer B. Srivastiva, V. Govind Manian, William Monroe and Christopher Potts. 2016. “Fitting In or Standing Out? The Tradeoffs of Structural and Cultural Embeddedness”. American Sociological Review 81(6): 1190-1222.