Browse for Publications by:

Browse for Journal Articles based on DHS data by:

orange publication summary banner small

Document Type
Methodological Reports
Publication Topic(s)
Anemia, Child Health, Maternal Health, Nutrition
Recommended Citation
Pullum, Thomas Pullum, Deborah Kortso Collison, Sorrel Namaste, and Dean Garrett. 2017. Hemoglobin Data in DHS Surveys: Intrinsic Variation and Measurement Error. DHS Methodological Reports No. 18. Rockville, Maryland, USA: ICF.
Download Citation
RIS format / Text format / Endnote format
Publication ID


Download this publication

Small PDF IconHemoglobin Data in DHS Surveys: Intrinsic Variation and Measurement Error (PDF, 2084K)
There is no printed copy available to order.


The accurate estimation of anemia is important for tracking and targeting public health interventions. The primary source of anemia data in low and middle-income countries is The Demographic and Health Surveys Program, in which hemoglobin concentration is assessed with a portable hemoglobinometer. This methodological report examines measurement error of hemoglobin assessment and the intrinsic variation in hemoglobin concentrations among children (age 6-59 months), nonpregnant women of reproductive age (age 15-49), and men (age 15 and above). A total of 80 surveys in The Demographic and Health Surveys Program conducted between 2000 and 2017 were selected, which results in a total of 1,247,942 hemoglobin observations included in this report (children n=405,731, women n=607,101, men n=235,110). Data quality was assessed by examining bias in the sub-sampling strategy, data completeness, and digit preference. Dispersion of the hemoglobin concentrations was also explored but it was difficult to determine whether the patterns observed are the result of measurement error or intrinsic variations. There was little bias found in the situations where hemoglobin measurements were only taken on a subsample of the population, although in a few surveys there was a slight bias by head of household education level, wealth, and urban/rural residence. There were very few values outside of the plausible ranges (mean percent ranged from 0.1 to 0.2% depending on the subpopulation) and only a small percent of data was missing (mean percent ranged from 4.5 to 15%). Digit preference was found to occur for the digits 0 (11.6% of surveys) and 2 (10.5%), and for the combination of digits 6, 7, 8, and 9 (22.1%). Standard deviations were outside the range of 1.1 to 1.5 in many surveys for children (46.3% with excluded implausible values versus 58.8% with included implausible values), women (70.8% with excluded implausible values vs. 81.5% with included implausible values), and men (96.3% with excluded implausible values vs. 96.3% with included implausible values). Data was not normally distributed in many of the surveys, especially among adults. Hemoglobin concentrations were higher in urban regions and wealthy populations, and in these groups there was less data dispersion, skewness, and kurtosis. In conclusion, our findings indicate that the overall quality of data is high on some measures, although there are exceptions, especially wide standard deviations. Disentangling measurement error from intrinsic variation is difficult. Future research is needed to establish standard parameters that assess measurement error in the assessment of hemoglobin and other biomarkers.