Purpose of the Survey and Reference Period

A questionnaire for the second round of the Social Impact of COVID-19 survey was conducted by the CSO from 10^th-17^th June. Most individuals selected to participate received an email from the CSO asking them to complete the questionnaire online, while some were contacted via telephone. The questionnaire asked for information on the following topics:

• Concerns about the easing of COVID-19 restrictions
• Changes in income, expenditure and spending of potential additional money since the introduction of COVID-19 restrictions
• Weight change during the period of COVID-19 restrictions
• Labour market activity and working arrangements of COVID-19
• Compliance with official advice

Sample Selection

This survey is the second in a series on the Social Impact of COVID-19. The sample was generated from Labour Force Survey (LFS) respondents that agreed to be contacted for further research and that had provided an email address and phone number. The Labour Force Survey is a 2-stage sample design stratified using Administrative County and the Pobal HP Deprivation Index. For further information please see Labour Force Survey.

The first Social Impact of COVID-19 survey was conducted in April 2020 and comprised of 4,033 respondents and non-respondents. The sample for this second survey in the series included the full April 2020 sample plus an additional 1,628 individuals selected from the LFS.

Data Collection

The sample selection methodology resulted in a sample of 5,566 individuals. 4,739 of these were issued the questionnaire via email. A further 827 were scheduled to be contacted via CATI (Computer Assisted Telephone Interviewing). Data collection was closed on Wednesday 17th June 2020, at which point the achieved sample was 1,693 individuals.

Sample Design

Timeliness was a key priority in this survey and therefore the sample and subsequent weighting process is one of convenience to some extent. Some consideration needs to be given to the potential impact of sample design on response rates and achieved sample:

• The original LFS sample from which this sample was selected was based on Census 2011 data and designed to represent the population then, so new additions to the population may not be fully represented.
• The sampling frame excludes people who do not live in private households, so may not cover some of those more likely to be negatively affected by the spread of the disease.
• The sampling frame was composed of individuals who had responded through 5 waves of the LFS survey, which could have introduced bias into the sample, as their characteristics may differ in some way to people more inclined to drop out of a longitudinal sample over time.
• The sample did not include people that did not provide an email address, which means members of the population less likely to have an email address, such as older people or people without internet access, were likely to be underrepresented.
• There is a mode effect, whereby the method of administration of the questionnaire can impact responses. For example, the presence of an interviewer in CATI can encourage a higher response rate compared to self-administered web questionnaires. On the other hand, sensitive questions may be more honestly answered without an interviewer present.
• The achieved sample distribution could also have been impacted by non-response bias. This is caused by the fact that some respondents might be more inclined to respond than others, and people who respond to these surveys often have different characteristics compared to non-respondents.

The weighting procedure outlined below was designed to adjust for possible bias in the achieved sample as much as possible.

Weighting

The following weighting process was devised to counteract some of the potential bias within the sample, and to make the final weighted sample distribution as representative as possible of the population.

Stage 1: Non-response

In the first stage of the weighting process, each person in the sample was given a weight of 1. We utilised the current LFS non-response adjustment process, in which a stepwise logistic regression was conducted based on census household-level data, to generate response propensities based upon the following characteristics:

• Personal characteristics (of head of household) – Sex, Marital status, Education, Nationality, PES, Age, Social class, Ethnicity
• Household characteristics – Area type, Dwelling type, Tenure, Number of persons in household, Number of cars owned by household, Number of rooms, Household PC, Urban/Rural, County

The sample is then grouped into strata based on propensity score, for which non-response adjustments were calculated and applied to each respondent.

Stage 2: Calibration

In stage 2, Q1 2020 LFS population estimates were used to benchmark the dataset across key characteristics for calibration. The non-response adjustments were inflated match overall the population total and then calibrated using CALMAR^[1], to ensure that weighted sample totals matched the Q1 2020 benchmark totals for a number of key characteristics such as gender, age, education level, region, urban/rural location, household composition.

Population distributions from LFS Q1 2020 (the most recent estimates available at the time of analysis) were compared with the sample distribution across key characteristics both before and after weighting (see table below).

[1] CALMAR is the statistical software developed by INSEE. Calmar is a SAS macro program that implements the calibration approach and adjusts weights assigned to individuals using auxiliary variables.

Non-response adjustment has been used to address some of the imbalances between the original sample design and the achieved sample distribution as much as possible, and the subsequent calibration adjusts to key population totals to try and match current population distributions. However, given the non-random nature of the final sample selected, it is unlikely that we can fully account fully for bias inherent in the final sample. For this reason, caution should be taken when attempting to make inferences to the entire population from these results.

Questionnaire

The questionnaire covered a number of topics such as concerns about the easing of COVID-19 restrictions, changes in income, expenditure and spending habits since the introduction of COVID-19 restrictions, labour market activity and working arrangements of COVID-19, weight change during the period and compliance with official advice.

Data analysis

As the sample was drawn respondents in the LFS, we were able to analyse the data using information collected during that survey regarding household and individual characteristics. However, it should be noted that while we have attempted to capture some of the changes that may have occurred between the LFS interview and this survey. In most cases we are assuming that there is no change in household characteristics, such as location and broad household composition.

Some key analysis variables within this publication are:

Household composition

For the purposes of deriving household composition, a child was defined as any member of the household aged 17 or under. Household were then categorised as:

• 1 adult households
• 2+ adult households
• Households with children

Highest level of education attained

This classification is derived from a single question and refers to educational standards that have been attained and can be compared in some measurable way and it is included in the core LFS on an ongoing basis.

The question is phrased as follows:

What is the highest level of education or training you have attained?

For the purposes of this publication these have been classified as follows:

• Higher secondary education or lower
• Post-secondary or Short-cycle tertiary
• Third-level bachelor degree or higher 

Marital status

Marital status refers to the current marital status of the respondent. In order to achieve appropriate sample sizes for each group. the responses were grouped as:

• Never Married
• Married
• Separated, divorced, or widowed

Tenure status

Tenure status refers to the nature of the accommodation in which the individual resides. The status is provided by the respondent of the household questionnaire during the interview and responses are classified into the following two categories:

• Owner-occupied
• Rented

Urban/rural location

Areas are classified as Urban or Rural based on the following population densities derived from Census of Population 2016:

Urban

Population density >100,000
Population density 50,000 – 99,999
Population density 20,000 – 49,999
Population density 10,000 – 19,999
Population density 5,000 – 9,999
Population density 1,000 – 4,999

Rural

Population density <199 – 999
Rural areas in counties

NUTS2 Regions

Regional analysis is presented in this publication are based on the NUTS2 (Nomenclature of Territorial Units) classification used by Eurostat. The regions are categorised as follows:

• Northern and Western
• Southern
• Eastern and Midland

These regions are comprised as follows:

Analysis by deprivation/affluence

The Pobal Haase-Pratschke Deprivation Index uses Census data to measure levels of disadvantage or affluence in a particular geographical area. More detailed information on the index can be found on the Pobal website.  Analysis based on deprivation index in this publication are presented using the following classification:

• Very disadvantaged
• Advantaged
• Average
• Affluent
• Very Affluent

Work situation

To capture respondents’ work situation during COVID-19 they were asked “What is your employment status?” with the following response options:

• Employee and currently engaged in work duties
• Employee pre COVID-19, currently not working, expecting to return to the same job
• Employee pre COVID-19, currently not working and not expecting to return to the same job
• Self employed and currently engaged in work duties
• Self employed but not carrying out work duties
• Unemployed pre COVID-19 and currently unemployed
• Other (e.g. retired, student, disabled)

The results from this question were combined to create a classification of work situation as follows:

• Employed
• Newly labour inactive
• Still unemployed
• Labour market inactive

Note that due to conceptual and methodological differences, references to “employment” and “unemployment” within this survey and the Labour Force Survey, results presented in this publication do not constitute official measures of employment (see the Labour Force Survey for further details on official employment statistics).

Acknowledgement

The Central Statistics Office wishes to thank the participants for their co-operation in agreeing to take part in the Social Impact of COVID-19 Survey and for facilitating the collection of the relevant data.

Go to next chapter >>> Contact Details