Sample size change in Survey data

[olabuh]

I am analyzing a survey data with a raw sample size of about 2.7 million, when I run the proc survey procedures, the sample size comes down to about 2 million and I am sure this is not explained by the missing data because I ran the proc freq and the missing data were very small and the sample size with the normal proc procedures was 2.7 million. Does anyone know why?

Thanks.

[Kurt_Bremser]

See if you can observe the effect with a small subset (~100 observations) of your data. If yes, post that here (in a data step with datalines, see my footnotes), and your code, so we can recreate the issue.

[PaigeMiller]

---

@olabuh wrote:

I am analyzing a survey data with a raw sample size of about 2.7 million, when I run the proc survey procedures, the sample size comes down to about 2 million and I am sure this is not explained by the missing data because I ran the proc freq and the missing data were very small and the sample size with the normal proc procedures was 2.7 million. Does anyone know why?

Thanks.

---

What specific analysis are you performing? What specific SAS PROCs are you using?

[Watts · Accepted solution]

If you're specifying design variables (STRATA and/or CLUSTER variables), maybe some of these values are missing. Observations with missing STRATA or CLUSTER values are excluded from the analysis unless you specify the MISSING option. Also, observations with missing or negative WEIGHT values are excluded. There's more information in the survey proc documentation section Missing Values.