High-Frequency Monitoring of COVID-19 Impacts on Households 2021-2022, Rounds 1-3
Socio-Economic/Monitoring Survey [hh/sems]
Round 1 of the High Frequency Survey was conducted from 2021-05-18 to 2021-06-16.
Round 2 was conducted from 2021-10-17 to 2021-11-01.
Round 3 was conducted from 2022-04-10 to 2022-05-15
The World Bank has launched a fast-deploying high-frequency phone-based survey of households to generate near real time insights into the socio-economic impact of COVID-19 on households which hence to be used to support evidence-based policy responses to the crisis. At a time when conventional modes of data collection are not feasible, this phone-based rapid data collection method offers a way to gather granular information on the transmission mechanisms of the crisis on the populations, to identify gaps in policy responses, and to generate insights to inform scaling up or redirection of resources as the crisis unfolds.
Unit of Analysis
This version contains data from Round 1, Round 2 and Round 3 that have been anonymized for public use.
The Malaysia High Frequency (HiFy) Phone Survey covers the following:
• Individual and household demographics
• Access to health services and vaccination
• Employment and income
• Food security
• Education and at-home learning
• Access to internet and digital services
• Household concerns
• Household coping strategies
• Safety nets
• Satisfaction with government’s COVID-19 response
Producers and sponsors
Global Tax Program Multi-Donor Trust Fund
A mobile frame was generated via random digit dialing (RDD), based on the National Numbering Plans from the Malaysian Communications and Multimedia Commission (MCMC). All possible subscriber combinations were generated in DRUID (D Force Sampling's Reactive User Interface Database), an SQL database interface which houses the complete sampling frame. From this database, complete random telephone numbers were sampled. For Round 1, a sample of 33,894 phone numbers were drawn (without replacement within the survey wave) from a total of 102,780,000 possible mobile numbers from more than 18 mobile providers in the sampling frame, which were not stratified. Once the sample was drawn in the form of replicates (subsamples) of n = 10.000, the numbers were filtered by D-Force Sampling using an auto-dialer to determine each numbers' working status. All numbers that yield a working call disposition for at least one of the two filtering attempts were then passed to the CATI center human interviewing team. Mobile devices were assumed to be personal, and therefore the person who answered the call was the selected respondent. Screening questions were used to ensure that the respondent was at least 18 years old and within the capacity of either contributing, making or with knowledge of household finances. Respondents who had participated in Round 1 were sampled for Round 2. Fresh respondents were introduced in Round 3 in addition to panel respondents from Round 2; fresh respondents in Round 3 were selected using the same procedure for sampling respondents in Round 1.
In Round 1, the survey successfully interviewed 2,210 individuals out of 33,894 sampled phone numbers. In Round 2, the survey successfully re-interviewed 1,047 individuals, recording a 47% response rate. In Round 3, the survey successfully re-interviewed 667 respondents who had been previously interviewed in Round 2, recording a 64% response rate. The panel respondents in Round 3 were added with 446 fresh respondents.
A full probability-based sampling methodology was used for each survey round; the complete weight is for the entire sample adjusted to 2019 and 2021 population estimates from the Malaysia Department of Statistics for each consecutive round, and excluded the portion of population that was not part of the target universe, particularly for age group and citizenship status.
The weighting scheme for Round 1 was developed with the following adjustments via weighting: 1. Base Weight: a base weight was calculated as the inverse of the probability of a number being dialed. 2. Non-response Weight: a non-response weighting adjustment was performed using a weighting-class adjustment by the inverse of Response Rate 3 (as defined by AAPOR) by the sample design stratum. Phone numbers that were removed via pulsing were considered Not Eligible for this weighting adjustment. 3. Raking and Trimming: a post-stratification weighting adjustment was performed using the aforementioned benchmark source for national population figures. An outlier analysis of the weights was then performed, with those beyond 3 standard deviations of the mean being trimmed. This was done as an iterative process (raking, trimming, and raking again) until weights were stable within a comfortable max weight. 4. Rescaled: weights were also delivered in a rescaled format. The weighting scheme for Round 2 was developed using adjustments in the form of raking and trimming, as well as weight rescaling.
There are two weights provided in Round 3: the cross-sectional weight and the panel weight. These two weights were developed using the same procedure in Round 2. The cross-sectional weight was assigned to both fresh and panel samples to ensure the representativeness of the Round 3's cross-sectional samples, while the panel weight was assigned only to panel respondents for longitudinal analysis.
Dates of Data Collection
Data Collection Mode
Computer Assisted Telephone Interview [cati]
The questionnaire is available in three languages, including English, Bahasa Melayu, and Mandarin Chinese. It can be downloaded from the Downloads section.
Estimates of Sampling Error
In Round 1, assuming a simple random sample, with p=0.5 and n=2,210 at the 95% CI level, yields a margin of sampling error (MOE) of 2.09 percentage points. Incorporating the design effect into this estimate yields a margin of sampling error of 2.65% percentage points.
In Round 2, the complete weight was for the entire sample adjusted to the 2021 population estimates from DOSM’s annual intercensal population projections. Assuming a simple random sample with p=0.5 and n=1,047 at the 95% CI level, yields a margin of sampling error (MOE) of 3.803 percentage points. Incorporating the design effect into this estimate yields a margin of sampling error of 3.54 percentage points.
Among both fresh and panel samples in Round 3, assuming a simple random sample, with p=0.5 and n=1,113 at the 95% CI level yields a margin of sampling error (MOE) of 2.94 percentage points. Incorporating the design effect into this estimate yields a margin of sampling error of 3.34 percentage points.
Among panel samples in Round 3, with p=0.5 and n=667 at the 95% CI level yields a margin of sampling error (MOE) of 3.80 percentage points. Incorporating the design effect into this estimate yields a margin of sampling error of 4.16 percentage points.
Use of the dataset must be acknowledged using a citation which would include:
- the Identification of the Primary Investigator
- the title of the survey (including country, acronym and year of implementation)
- the survey reference number
- the source and date of download
World Bank. Malaysia - High-Frequency Monitoring of COVID-19 Impacts on Households 2021-2022, Rounds 1-3 (MYS HiFy 2022-2023). Ref: MYS_2021-2022_HFS_v01_M. Downloaded from [uri] on [date].
Disclaimer and copyrights
The user of the data acknowledges that the original collector of the data, the authorized distributor of the data, and the relevant funding agency bear no responsibility for use of the data or for interpretations or inferences based upon such uses.