Why Does the Thai Government Struggle to Retain Scholarship Recipients?

Why does the Thai government have trouble retaining scholarship recipients (setting aside feelings)?

The issue of government scholarship students has come back lately, so I’d like to share my side as a MHESI scholarship recipient. At first I wanted to share from the perspective of someone working off a scholarship, but I felt many people had already shared that, and I don’t want to cheer anyone on to accept or refuse a scholarship, because each side is going at it fiercely, haha.

I’ve read many posts, and I see proposals for what the state should do: raise salaries? Do talent mobility? But we really don’t know whether they’d work. So today I’d like to share another angle, that of a physicist who does a little research, using an agent-based model (ABM) to test which policies affect keeping government scholarship recipients working until they fulfil their obligations. (The post is a bit long. The conclusion is in the last paragraphs.)

An ABM builds many simulated agents and observes the collective behavior of the system. It’s usually used to study things like retention rates that arise from individual-level decisions. In my experiment, each agent represents one scholarship recipient. Each has a field of study (STEM, medicine, law, social sciences, education), a scholarship duration and value, a cost of living depending on the destination country, and a preference offset toward working in the public sector. Every year during the service period, each agent compares the utility of four options: (1) continue serving, (2) leave and pay the penalty themselves (voluntary exit), (3) leave with a private employer paying the penalty for them (employer buyout), and (4) leave under a talent mobility mechanism, in which part of their income is returned to the public sector. Each option is calculated as the present value of net income over the working life, with a 5% discount rate.

Here is what I found.

1. The penalty mechanism has no effect on retention.

No agent left by paying the penalty out of their own pocket, across every penalty level tested (from 1.5 times to 5 times the scholarship value). The reason is a structural liquidity constraint. A recipient who returns to government service at 30,000 baht/month has a maximum fundraising capacity, from accumulated savings and personal loans combined, of about 1.9 million baht, while the penalty for a typical PhD scholarship at 3 times the amount received is about 27 million baht. The gap is roughly 14 times.

This finding suggests the penalty mechanism was designed on assumptions that don’t match reality: the penalty is set according to the cost of the scholarship, not the civil servant’s ability to pay. A policy designed to deter individual decisions therefore has no channel through which it can actually take effect. An interesting by-product is that raising the penalty increases government revenue, but not because recipients pay it themselves. It is because private employers pay higher penalties as the price of buying them out. So the relationship between revenue and retention is a perverse trade-off: the configuration where the state earns the most is the one where the most people leave.

2. Employer buyout is the only exit channel that actually works.

When employer buyout is switched off entirely, retention rises to 100% under every condition tested. This means all of the variation in retention in the system comes purely from the demand side, the frequency and size of the offers sent by private companies, not from the decisions of individual recipients.

The distribution of agents who get bought out reflects the structure of private sector demand, not population shares. The STEM group is 35% of the whole population but 59% of those who leave. The medicine group is 16% of the population but 21% of those who leave.

3. The service multiplier, or length of the service period, works through the opposite mechanism from what the policy intends.

Extending the compulsory service period reduces retention from 84% at 1x to 66% at 3x when buyouts are allowed. Each additional year a recipient stays in the system is one more year in which private employers have a chance to send an offer. A policy designed to lock recipients in the system longer may therefore produce the opposite of the intended result.

4. Raising salaries is effective only when paired with talent mobility.

(Talent mobility here means being seconded to work in the private sector, with the private sector paying compensation back to the home agency.)

Raising the salary of scholarship holders from 360,000 baht/year to 720,000 baht/year (a 100% increase) has no effect on retention through the buyout channel (73.2% vs 72.0%), because the buyout salary equals the private salary x 1.30 (a value I set), and the gap with the private sector, especially in the STEM and medicine groups with high private salary multipliers, still exceeds the raise.

However, a raise of the same size increases retention by about 10% when buyouts are replaced by the talent mobility mechanism (76.4% to 86.0%). The mechanism behind the difference is that talent mobility reduces the private sector’s advantage to a level where a salary adjustment can actually motivate people. (With a payback rate of 20%, net income during mobility for STEM is about 720,000 baht/year, which matches the government salary level in the scenario tested.)

5. Talent mobility offers a different value proposition (and may not always be the better option).

Under the pure talent mobility mechanism (no buyouts), effective retention rises to 100%, because everyone who leaves is still under a payback agreement to their home agency and hasn’t disappeared from the system entirely. However, strict retention (people who truly serve out the obligation) is 76.4%, slightly higher than under buyouts (73.2%), because the mobility offer has lower net income than a buyout, so recipients at the margin of the decision still choose to continue serving.

On revenue: baseline mobility generates about 123 million baht for the state, compared with 3,596 million baht from buyout penalties, a difference of about 29 times. This difference comes from the structure of the numbers, not from the efficiency of the policy. The lump-sum penalty per person is about 5.9 million baht, while the value of the 20% payback for STEM is about 1.4 million baht. Mobility suits objectives of sector connection and talent re-circulation more than revenue maximization.

It’s a bit long, and may need a few reads, haha (if anything is confusing, leave a comment).

Overall, my policy conclusion is this.

The current policy framework may be misdiagnosing the problem. Retention is determined by the demand side (the behavior of private employers), not by the individual choices of recipients. But all the policy instruments in place are designed to change individual incentives.

What would likely be most effective is something not yet in the current framework: legal or contractual measures that limit private employers’ ability to pay the penalty on a recipient’s behalf. The model predicts this would push retention back to 100%.

However, the model doesn’t simulate the market response that would follow, such as benefits being transferred through other channels, or employers waiting until the obligation is over before hiring, which could make the real result lower than the model predicts.

These findings also point to something to consider in measuring policy. As long as retention is counted as a binary variable (served the full term or not), the system can’t see the difference between people who leave entirely and people who leave under an arrangement that still carries an obligation to the public sector. Reporting effective retention separately from strict retention (I’m not sure whether this data is collected) would give a more accurate picture for policy decisions.

This post is a preliminary summary. The full model specification, parameter calibration, and complete sweep results are being written up, and I’ll attach a GitHub link for anyone who wants to look further.

Pai