Problem 🎯

An interviewer at Meta asked:

How would you measure the effect of parents being on Facebook on teenagers?

Tips💡

Defining “effects” of parents being on FacebookThink about the various ways parents might influence teenagers on Facebook. This could range from content creation and consumption to privacy practices. Can you define ‘effect’ more specifically?
Discuss data collection & preparationConsider the types of data that might be available on Facebook that can provide insights into teenager behavior and parental influence. What kind of features would be informative for this analysis?
Conduct regression analysisThis is not an AB testing framework as you cannot enforce a group of teenagers to be connected to their parents on Facebook. You need to use historical data and conduct a regression analysis that measures the effects of parents on teenagers.

Solution ✍️

Interviewer: Let’s jump into a case question. Suppose you are tasked with measuring the effect of parents on teenagers on Facebook. How would you approach this problem?

Candidate: First, it’s crucial to define the scope of ‘effect’ in this context. Are we considering the impact on content consumption, creation, social behaviors, or privacy settings of teenagers?

Interviewer: Great start. Let’s say we are focusing on content creation behaviors such as posting and sharing.

Candidate: Also, do I have access to historical data for this analysis? I presume that we need to study naturalistic behavior, and that we cannot run an AB experiment. We can’t deliberately force a group of teenagers to be connected to Parents.

Interviewer: Yes, let’s say that you have each user’s profile, connection, activity data for the past two years. How would you use this?

Candidate: Understood. Firstly, I’d collect historical data on teenagers’ content creation behaviors, like the frequency of posts, likes, shares, comments, and privacy setting adjustments. I’d also gather data on the types of content consumed and the duration and times of their active sessions. Ultimately, I would aggregate the use data at the weekly level. This would help me change how the behavior of a teenager changed over time.

User	Week	Average_Session_Time	Num_of_Shares	Number_of_Posts
xyz	1	8	4	2
xyz	2	10	3	5
xyz	3	3	16	12
xyz	…	…	…	…
abc	1	4	4	3
abc	…	…	…	…

Commentary: The candidate should have also discussed how to identify whether the users are “teenagers”. The age group among teenagers are 13 to 18. And, this is defined at the time of sign-up as a user needs to specify the birth date.

Interviewer: And how would you consider the parent’s activity in this?

Candidate: I’d create a binary variable, named Parent_Active, which indicates whether a parent is active on Facebook (1 for yes, 0 for no). I am sure that Facebook readily has information on whether a user is the parent of a teenager.

User	Week	Average_Session_Time	Num_of_Shares	Number_of_Posts	Parents_Active
xyz	1	8	4	2	0
xyz	2	10	3	5	0
xyz	3	3	16	12	1
xyz	…	…	…	…	…
abc	1	4	4	3	1
abc	…	…	…	…	…

Commentary: On Facebook, it’s not always explicitly defined whether two users are family members, and whether a user A is the parent of a user B. The candidate should have known that defining the nature of connection is optional, and that only a fraction of users do this. You would need to build a classification model that predicts whether two users are a pair of parent and teenager using variables including ip_address, profile_images, age_of_teenager, age_of_parents and e.t.c. You can then use the output from the classification model to map which teenager is connected to a parent.

Interviewer: Interesting. How would you quantify the impact of parents on teenagers’ posting behavior?

Candidate: I’d set up a regression analysis to quantify this impact. For instance, the dependent variable could be Number_of_Posts, and I’d use Parent_Active, Average_Session_Time, and Number_of_Family_Friends as independent variables. The regression equation would look like:

Number_of_Posts = B0 + B1 * Parent_Active + B2 * Average_Session_Time + B3 * Number_of_Family_Friends

This approach can help us understand the influence of having an active parent on Facebook on a teenager’s posting behavior. I could also consider other covariates to include to better explain the model. Ultimately, I could use cross-validation and R-squared adjusted to determine whether I have a well-fitted model.

Interviewer: How would you then interpret the findings from this regression analysis?

Candidate: I’d closely look at the coefficient of the Parent_Active variable. If it’s significant and negative, it would suggest that having parents on Facebook decreases the number of posts by teenagers. Conversely, a significant and positive coefficient would indicate an increase in posting behavior. If the coefficient is not statistically significant, it would be hard to confidently assert that parents’ presence on Facebook influences teenagers’ posting behavior.

Assessment 📝

1. Business Sense & Insights

Rating: Proficient

The candidate demonstrates a good understanding of the business context by acknowledging the importance of understanding the effect of parents on teenagers on Facebook. His attempt to derive insights related to teenagers’ content creation behavior on the platform in response to parental activity. However, the candidate could improve by better aligning the insights with specific business goals of Facebook, such as user engagement, user retention, or content quality.

2. Communication & Presentation Skills

Rating: Proficient

The candidate communicates their thought process and methodology clearly, step by step. He asks clarifying questions to ensure understanding of the problem and use tables to communicate how data might be structured. However, while he is mostly effective in their communication, he could improve by avoiding assumptions (e.g., regarding the availability of parent-teenager relationship data) and explaining their thought process and the implications of their findings in a way that is more accessible to non-technical stakeholders.

3. Technical Methodology

Rating: Borderline

Justification: The candidate shows an understanding of regression analysis and the need for feature engineering but overlooks the complexities in defining and extracting the parent-teenager relationships from the data. While he does suggest creating a classification model to predict parent-teen relationships, he doesn’t elaborate on how this would be done, what features would be used, or how the model would be evaluated and validated. He also doesn’t discuss how to handle issues like multicollinearity among independent variables, non-linearity, or overfitting in the regression model.

4. Problem Solving & Critical Thinking

Rating: Proficient

The candidate takes a structured approach to the problem by first clarifying assumptions and then outlining the steps he would take to analyze the data. He demonstrates critical thinking by considering the importance of data aggregation and the need to adjust for covariates in the regression model. However, he could improve by thinking more deeply about potential challenges and limitations of their approach. For instance, regression requires assumptions (like linearity, independence, etc.) that might not always be met. Acknowledging these assumptions and how he plans to test and address them is crucial.

[Meta] Parents on Facebook

Problem 🎯

Tips💡

Unlock now by joining.