Study healthcare workforce mobility without confusing nationality and migration

1. Replace “workforce migration” with a measurable question

Choose between questions such as the training background of active nurses on one date, the number of people entering an employer during a year, and the number moving across a national border. These require different records. A new hire might have changed hospitals within the same city; a foreign national might have lived locally for decades.

Write the intended unit, profession, employer coverage, geography and observation window before requesting data. For a staffing description, specify whether the unit is a person, a position or a full-time equivalent. For a movement question, define the origin, destination and event date. An unfilled position belongs in a vacancy analysis, not in a count of active people.

2. Keep background characteristics separate from movement

The WHO’s 2023 mobility report overview distinguishes foreign-trained and foreign-born workers within the active workforce. Those categories describe different characteristics. The report brings together several data mechanisms and much of its underlying availability concerns 2011–2021; its publication date does not make those observations current.

For your study, record citizenship, birthplace and country of qualifying training in separate fields if they are available and relevant. Define which qualification counts when a worker has several. Do not derive one field from a name, language or another field. If the source records citizenship only, a question about training location remains unanswered.

Reference: WHO: report on global health worker mobility (2023), publication overview

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

3. Build a source sheet before a comparison table

For each candidate dataset, retain the issuing body, exact table or file, version, observation dates, retrieval date, definitions and exclusions. Ask whether private employers, temporary staff and workers outside clinical practice are included. A table headed “health professionals” cannot be assumed to have the same coverage as one headed “licensed nurses”.

Treat a regional total as a separate analytical product that needs compatible definitions. A Saudi table supplies evidence about its defined Saudi population; it does not fill missing observations for other GCC countries. If no suitable current source is available, narrow the question or document a data request. Do not substitute a policy target for an observed count.

4. Diagnose an invented two-site mismatch

Imagine Site A supplies an invented census of 120 active nurses on 30 June. Training location is recorded for 84 people, of whom 42 qualified outside the study country; it is missing for 36. These are teaching numbers with no connection to a real employer. Site B supplies 80 funded nursing positions, including vacancies, and has no person-level training field.

The two totals cannot answer the same workforce-composition question. Adding 120 people to 80 positions would mix units. Keep Site B out of that comparison until an appropriate person-based source is obtained. Neither file contains an origin–destination movement history, so neither supplies an annual immigration count.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

5. Show how the denominator changes the answer

Within the fictional Site A records with known training location, the foreign-trained share is 42 ÷ 84 = 50%. Across all 120 people, 42 are known to be foreign-trained, equivalent to 35% of the total. These statements have different denominators and must carry their labels. Coding the 36 unknown values as locally trained would conceal uncertainty.

If all 36 unknown values were locally trained, the full-workforce share would be 35%; if all were foreign-trained, it would be 78 ÷ 120 = 65%. This 35%–65% range is a simple missing-data bound under the invented premises. It is not a confidence interval, an adjusted estimate or a claim that either extreme is likely. Ask how the missing records arose before proposing a more elaborate analysis.

6. Write a small analysis plan that can be checked

The proposed plan is to describe Site A only, at the named census date, using the supplied active-worker definition. First check duplicate people, eligibility and mutually exclusive training categories. Then report the total, known and unknown counts together. Preserve the dictionary and any later change to its definition so that the published wording can be traced to the analysis.

A later second census would allow a comparison of two stocks if definitions remain compatible. Its difference alone would not identify arrivals: hiring, exits, classification changes and coverage changes may all affect the count. Study movement with appropriate event records, and study reasons with a design capable of addressing that question. The current example has neither.

7. Draft a bounded result, then identify what is missing

A defensible teaching sentence is: “In the fictional Site A census, training location was available for 84 of 120 active nurses; 42 of those 84 qualified outside the study country.” Follow it with the missing-data count and the source definition. Do not shorten it to “half of nurses migrated”, which changes both the population and the concept.

Before working with an actual workforce file, establish permitted access, necessary fields and disclosure rules with the responsible institution. A public descriptive table and an identifiable employment record are different data products. This guide does not grant access, approve a study or determine a legal basis for processing individual records.

Questions for a practical final check

• Name people, positions or full-time equivalents explicitly.
• Separate citizenship, birthplace, training and movement.
• Retain exact dates, definitions, source version and exclusions.
• Report missing values with the denominator.
• Avoid pooling incompatible sites or countries.
• Keep descriptive results separate from causal or policy claims.