All notes
Economics

Average 7.3 million, median 4.2 million: Uzbekistan publishes its first median wage

Uzbekistan's statistics committee has published a median wage for the first time. What the number shows, why it can't be compared directly with the average, and what it has in common with p50 latency.

On 27 October, Uzbekistan's National Statistics Committee published its regular wage release. This time it had a new section: the median wage. It is the first time the committee has published this figure. I work on backends and payment systems and look at numbers all day, so this one caught my attention.

What happened

According to the press release, the average monthly nominal accrued wage for January–September 2026 was UZS 7,274.1 thousand, up 18.0% on the same period last year. The committee calculates it excluding agricultural enterprises and small businesses.

The same document gives, for the first time, a median wage: UZS 4,166.3 thousand for January–August 2026. The median is the amount that half of employees earn less than and half earn more than.

By region, the highest median is in Tashkent city (UZS 6,791.2 thousand) and the lowest is in Syrdarya region (UZS 2,998.4 thousand). By sector, banking, insurance, leasing and credit leads (UZS 11,363.9 thousand), and accommodation and food services is at the bottom (UZS 2,860.0 thousand).

Sector Average, Jan–Sep Median, Jan–Aug
Uzbekistan 7,274.1 4,166.3
Banking, insurance, leasing, credit 19,712.6 11,363.9
Information and communication 17,006.9 8,752.4
Transport and storage 11,269.2 8,300.5
Industry 8,365.9 5,858.5
Trade 7,694.4 4,654.9
Education 5,015.2 3,207.6
Health and social services 4,454.2 3,267.0
Accommodation and food 5,359.2 2,860.0

UZS thousand. Source: National Statistics Committee.

One important caveat. As the committee itself points out, the median covers January–August because the person-level administrative data for September is not yet complete. So it cannot be compared directly with the January–September average. I put them side by side only to show how wide the gap is.

Context

The average wage is the total wage bill divided by the number of employees, so a small group of very high earners pulls it up. Gazeta.uz quotes the committee's simple example: for five salaries of 3, 4, 5, 6 and 22 million soums, the average is 8 million and the median is 5 million. The committee itself notes that the average wage "does not mean that most employees earn exactly that amount".

Until now, public discussion mostly revolved around the average, and many people were puzzled when they compared "7 million on average" with their own pay. The median answers that question far more honestly.

A few methodology details from the release are worth knowing:

  • The median is calculated from the administrative database of "income in the form of wages", before personal income tax.
  • Employees of organisations with branches are assigned to the legal entity's registered address. That may push Tashkent figures up somewhat, since many head offices are in the capital.
  • The calculations are experimental: the indicator is not yet part of the annual state programme of official statistics.

What it means

For people. The worry "I earn less than the average" often comes from the wrong comparison. Against the median, the picture is clearer.

For businesses. For pay policy, job offers and cost planning, the sector and regional median is a better benchmark than the average, especially in IT and finance, where the gap between the two is large.

For engineers. This is where it gets interesting for me.

  1. It's the same story as p50. Every backend developer knows that average latency hides a lot, which is why we look at p50, p95 and p99. Economic statistics follow the same logic.
  2. You can't compute a median from aggregates. The average is simple: the wage bill divided by average headcount, which you can get from companies' summary reports. A median needs a row per person, which is why the committee relied on person-level administrative data. Another subtle point: averaging regional medians does not give you the national median. Many systems make the same mistake when they average per-service p95s and call it the "overall p95". At scale this is usually solved by keeping the full distribution or by using mergeable sketches such as t-digest.
  3. It has direct value for fintech. If credit scoring, affordability checks and limit setting use the average wage as "typical customer income", the model becomes too optimistic. A regional and sector median is a more sensible baseline. Of course, it does not replace verifying a customer's actual income.

My take

I think this is the right and necessary step. The committee chose the number that is more accurate, not the one that looks nicer. Labelling the work as experimental and spelling out the limitations in the release itself also builds trust.

That said, I have a few wishes for what comes next:

  • Distribution. One median is good, but deciles (10%, 25%, 75%, 90%) would show inequality much more precisely.
  • Same period. Publishing the average and the median for the same period would make comparison far easier.
  • Machine-readable format. Right now the numbers live in a PDF. A CSV or an open API would let analysts and developers use them much more.
  • Coverage. Small businesses and agriculture are not yet included. Adding them would complete the picture.

Overall, this figure is the first official look at what the labour market looks like for an "ordinary employee". Wage discussions should now be built around it.

Sources: National Statistics Committee of Uzbekistan, press release, 27.10.2026; stat.uz news item; Gazeta.uz. This is personal opinion, not investment advice.

// comments

Loading comments...

    // next step

    Let's build something together.

    Message me on Telegram
    aslbek@asldev: ~
    ❯