Wait, what? It’s neither one-shot nor prisoner’s dilemma.
Hiring, like almost all real-life interactions, has many sub-games, some of which are zero-sum and some of which are positive sum. In fact, the main payout for both sides is the opportunity for many positive-sum (and some zero-sum) games in the future of an employment relationship. Also, there is legal recourse for the more egregious kinds of defection.
I’ve never had any future employment-related interactions with a company to which I have applied but not been hired to work for. I am confident Zvi will have no future interactions with the people whose applications he rejected. How is that not a one-shot? Intuitively I would be inclined to say that if both sides cooperate, then the sequence of games would continue; it looks to me like ‘cooperate’ basically means ‘commit to more games’. I’m happy to be wrong though, and if there’s a solid examination of the subject you can recommend I will certainly check it out.
The players (presumably) expect that there’s a chance of future interactions, even if that chance does not obtain. Zvi will have no future interaction with the un-hired applicants, but neither party knew that when the interaction started. It turns out to be one-shot in those cases, but it wasn’t known to be one-shot in advance and there are some outcomes which are not one-shot.
More importantly, the payout matrix doesn’t match the PD pattern—defecting does not pay more than cooperating. Actually, I’m not sure exactly what actions you’re mapping to cooperate/defect when you assert PD-like qualities. Is cooperating “try to get hired” and defecting “don’t answer your messages”? Or something else?
Is cooperating “try to get hired” and defecting “don’t answer your messages”? Or something else?
Good questions—the actual reasoning I used went approximately like this:
1. Hirers (Zvi) and applicants (everyone else) both complain about how terrible the process is.
2. Behavior from the other side you hate is probably defection.
3. This means both sides almost always defect.
4. PD is the famous case where both sides always defect. This is probably some form of PD.
I didn’t actually assign what the payoffs were, but while reading Zvi’s complaints I was focused largely on the question of courtesy, like phone etiquette and ghosting. So I guess the best approximation of my feelings would be to model cooperate as be courteous and defect as do not be courteous.
Courtesy takes effort and attention; ignoring it does not; therefore defect yields a better payoff. Both parties prefer to be treated well by the other side without exerting any effort themselves. I don’t see any motivation to treat each round as a different game because there’s no reason to expect that people will vary their choices—I expect ghosty people to ghost all the way through, and courteous people to be courteous all the way through. Being discourteous is in this way not meaningfully different from a prisoner ratting on his colleague over three interviews with police rather than in one interview.
I experienced a C-C outcome by those lights that still was not me getting hired; the company in question kept me informed about the state of the process, and followed up with me after another candidate was selected. This is not a predictable scenario though, because it was driven by their expectation of being able to hire two people, and then learning after I interviewed they could only hire one.
Ah, I think I see. The comfort/pleasantness aspect has a PD-like payoff (defecting saves effort, but hurts the other participant). I’d argue that the fact that D-C (you defect, opponent cooperates) is smaller payout for you than C-C (you both cooperate) makes most insights from PD not applicable here.
I do an annoying amount of interviewing and hiring for my employer (which I do not represent and for whom I do not speak), and while we do somewhat often miss on candidate experience simply due to scale and internal miscommunication, we STRONGLY expect that even for no-hire cases there will be future interactions with candidates (for different roles in our company and as the candidate grows over time), and with their friends/family/coworkers. This leads us to absolutely expect that interactions are repeatable rather than one-shot.
Candidates likewise often realize that even if they don’t get this job today, they may want to apply to a different or future opening. Also, people move around to different companies, so even if it’s a different employer, you might meet the same person. That probably doesn’t hold for extremely tiny employers for high-turnover low-skill positions, but each employment world is shockingly small, and reputation has more pathways than one might naively think.
Agreed with this to the point that I think hiring is a pretty excellent example of how interactions that look one-shot actually turn out to be iterated.
(I’m also not sure whether iterated PD is meant to refer to “repeated, and you explicitly have a reputation” vs “repeated, and even without reputation explicitly taken into account there are ramifications and selection pressures of what sort of strategies survive over the long term.” I was under the impression that the latter was included in the term)
I’d argue that the fact that D-C (you defect, opponent cooperates) is smaller payout for you than C-C (you both cooperate) makes most insights from PD not applicable here.
Oh, that’s a very good point! I had put the weight on the expected outcome of D-D, but nobody chooses the outcome they want, only the move that seems to have the best payoff. I will dedicate some time to adjusting my intuitions on this, because I suppose the next natural question is, if C-C is higher payout than D-C, why does it so often seem to end up that way?
Based on your experience, I guess this to vary heavily by industry; skilled labor and professions have a much smaller hiring/applicant pool, and often have communities set up around their work besides. By contrast, the number of hirers/applicants for unskilled labor is much larger, and there usually aren’t communities built up around it. On the other hand, I would also expect unskilled labor to be more closely tied to the actual physical communities in which people live, on both the hirer and applicant sides. On still a third hand, the job-related reputation seems fairly low priority in my physical community. Hmm.
I’m also reminded that the payoffs (or costs) aren’t equal in real-life games. It is a lot more effort to send “Thank you but we are moving forward with other candidates” emails to 99 people than it is to send “Thank you but I have accepted another offer” emails to 2 or 3 hirers. Also, the effort involved in correctly formatting applications and avoiding grammar/spelling errors is much higher for poorly educated people than it is for me.
But the immediate takeaway is, I gotta shift my focus from D-D outcome to D-C move at the intuition level.
Wait, what? It’s neither one-shot nor prisoner’s dilemma.
Hiring, like almost all real-life interactions, has many sub-games, some of which are zero-sum and some of which are positive sum. In fact, the main payout for both sides is the opportunity for many positive-sum (and some zero-sum) games in the future of an employment relationship. Also, there is legal recourse for the more egregious kinds of defection.
I’ve never had any future employment-related interactions with a company to which I have applied but not been hired to work for. I am confident Zvi will have no future interactions with the people whose applications he rejected. How is that not a one-shot? Intuitively I would be inclined to say that if both sides cooperate, then the sequence of games would continue; it looks to me like ‘cooperate’ basically means ‘commit to more games’. I’m happy to be wrong though, and if there’s a solid examination of the subject you can recommend I will certainly check it out.
The players (presumably) expect that there’s a chance of future interactions, even if that chance does not obtain. Zvi will have no future interaction with the un-hired applicants, but neither party knew that when the interaction started. It turns out to be one-shot in those cases, but it wasn’t known to be one-shot in advance and there are some outcomes which are not one-shot.
More importantly, the payout matrix doesn’t match the PD pattern—defecting does not pay more than cooperating. Actually, I’m not sure exactly what actions you’re mapping to cooperate/defect when you assert PD-like qualities. Is cooperating “try to get hired” and defecting “don’t answer your messages”? Or something else?
Good questions—the actual reasoning I used went approximately like this:
1. Hirers (Zvi) and applicants (everyone else) both complain about how terrible the process is.
2. Behavior from the other side you hate is probably defection.
3. This means both sides almost always defect.
4. PD is the famous case where both sides always defect. This is probably some form of PD.
I didn’t actually assign what the payoffs were, but while reading Zvi’s complaints I was focused largely on the question of courtesy, like phone etiquette and ghosting. So I guess the best approximation of my feelings would be to model cooperate as be courteous and defect as do not be courteous.
Courtesy takes effort and attention; ignoring it does not; therefore defect yields a better payoff. Both parties prefer to be treated well by the other side without exerting any effort themselves. I don’t see any motivation to treat each round as a different game because there’s no reason to expect that people will vary their choices—I expect ghosty people to ghost all the way through, and courteous people to be courteous all the way through. Being discourteous is in this way not meaningfully different from a prisoner ratting on his colleague over three interviews with police rather than in one interview.
I experienced a C-C outcome by those lights that still was not me getting hired; the company in question kept me informed about the state of the process, and followed up with me after another candidate was selected. This is not a predictable scenario though, because it was driven by their expectation of being able to hire two people, and then learning after I interviewed they could only hire one.
Ah, I think I see. The comfort/pleasantness aspect has a PD-like payoff (defecting saves effort, but hurts the other participant). I’d argue that the fact that D-C (you defect, opponent cooperates) is smaller payout for you than C-C (you both cooperate) makes most insights from PD not applicable here.
I do an annoying amount of interviewing and hiring for my employer (which I do not represent and for whom I do not speak), and while we do somewhat often miss on candidate experience simply due to scale and internal miscommunication, we STRONGLY expect that even for no-hire cases there will be future interactions with candidates (for different roles in our company and as the candidate grows over time), and with their friends/family/coworkers. This leads us to absolutely expect that interactions are repeatable rather than one-shot.
Candidates likewise often realize that even if they don’t get this job today, they may want to apply to a different or future opening. Also, people move around to different companies, so even if it’s a different employer, you might meet the same person. That probably doesn’t hold for extremely tiny employers for high-turnover low-skill positions, but each employment world is shockingly small, and reputation has more pathways than one might naively think.
Agreed with this to the point that I think hiring is a pretty excellent example of how interactions that look one-shot actually turn out to be iterated.
(I’m also not sure whether iterated PD is meant to refer to “repeated, and you explicitly have a reputation” vs “repeated, and even without reputation explicitly taken into account there are ramifications and selection pressures of what sort of strategies survive over the long term.” I was under the impression that the latter was included in the term)
Oh, that’s a very good point! I had put the weight on the expected outcome of D-D, but nobody chooses the outcome they want, only the move that seems to have the best payoff. I will dedicate some time to adjusting my intuitions on this, because I suppose the next natural question is, if C-C is higher payout than D-C, why does it so often seem to end up that way?
Based on your experience, I guess this to vary heavily by industry; skilled labor and professions have a much smaller hiring/applicant pool, and often have communities set up around their work besides. By contrast, the number of hirers/applicants for unskilled labor is much larger, and there usually aren’t communities built up around it. On the other hand, I would also expect unskilled labor to be more closely tied to the actual physical communities in which people live, on both the hirer and applicant sides. On still a third hand, the job-related reputation seems fairly low priority in my physical community. Hmm.
I’m also reminded that the payoffs (or costs) aren’t equal in real-life games. It is a lot more effort to send “Thank you but we are moving forward with other candidates” emails to 99 people than it is to send “Thank you but I have accepted another offer” emails to 2 or 3 hirers. Also, the effort involved in correctly formatting applications and avoiding grammar/spelling errors is much higher for poorly educated people than it is for me.
But the immediate takeaway is, I gotta shift my focus from D-D outcome to D-C move at the intuition level.