Orientation Observation In-depth interviews Document analysis and semiology Conversation and discourse analysis Secondary Data Surveys Experiments Ethics Research outcomes



Social Research Glossary

About Researching the Real World



© Lee Harvey 2012–2017

Page updated 4 June, 2017

Citation reference: Harvey, L., 2012–2017, Researching the Real World, available at
All rights belong to author.


A Guide to Methodology

7. Secondary data
7.1 Introduction to secondary analysis
7.2 Extent of re-analysis of secondary data

7.3 Nature of the data

7.4 Data sources

7.5 Examining data sources
7.6 Methodological approaches

7.6.1 Positivism and secondary analysis
7.6.2.Phenomenology and secondary analysis
7.6.3.Critical approach to secondary analysis

7.7 Summary and conclusion

7.6 Methodological approaches
Besides these practical and conceptual issues, the use of published statistics also raises epistemological issues for social researchers.


7.6.1 Positivism and secondary analysis
The positivist approach, as explained in Section 2, is to adopt a natural sciences model and attempt theoretical explanations of apparent social phenomena. This, positivists do, in the main, by processes that allow the identification and measurement of social facts, with the intention of building cause-and-effect relations. Thus positivists find no fundamental epistemological objection to using government statistics (see Section

Positivists are concerned that all statistics are collected in systematic ways that make them valid and reliable. They are, therefore, critical of definitions and collection procedures. In this respect, while official statistics may not be based on sociological definitions, the Government Statistical Service is as scrupulous as it can be in documenting changes in definitional categories, difficulties in collection procedures, sample sizes and sampling frames. It thus provides users with clear information by which to judge the statistics collected. Positivists thus see official statistics as a useful source, which provide a much wider coverage than any independent enquiry could.

Positivists focus mainly on the reliability and consistency and accuracy of such statistical databases. Validity is given less scrutiny, there being an implicit assumption that the official definition of the statistic coincides with the conceptualisation of the concept and thereby validates the measurement. Positivists tend to be less inclined to examine the socio-political or historical context in which the data is collected and reported.

An example of the concern with consistency and validity is the well-recorded and researched problematic about the measurement of poverty. The two most prominent measures of material poverty in Europe are low income and material deprivation but these measures identify substantially different people as being poor. Hick (2015) used data from the British Household Panel Survey to assess the problem. The British Household Panel Survey interviewed adult members (aged 16 and over) living in sampled households on an annual basis. The survey commenced in 1991 and in 2009–10 was subsumed into the larger Understanding Society survey. Hick's analysis showed that although low income and material deprivation identify very different individual households as being poor they do however identify the same groups at being at risk of material poverty.

An example of a large-scale positivist secondary analysis of a statistical database is the study by a Swedish team comprising researchers at Karolinska Institutet, Linköping University, Stockholm South General Hospital and Karolinska University Hospital that analysed data from 10,000 patients from the large Swedish Heart Failure Registry (Thorvaldsen et al., 2013).

The research showed that patients with heart failure have high mortality. Heart failure affects 2–3% of the overall population and over 10% of the elderly worldwide, and is associated with high risk for early death and reduced quality of life. The researchers showed that early death in these patients was related to heart failure (rather than for example age) suggesting that better treatment for heart failure would reduce mortality.

Second, the researchers defined five risk factors for mortality: poor pump capacity, poor kidney function, low blood count and absent treatment with the drugs ACE inhibitors and beta blockers. If any one of these risk factors was present in a given patient, the mortality was so high that the patient would potentially benefit from a heart failure pacemaker, heart pump and heart transplantation. This risk persisted after adjustment for a large number of other factors, such as patients’ age, general health and other factors.

Drug therapy improves symptoms and reduces mortality and is well used. However, modern heart failure pacemakers, heart pumps and also heart transplantation are of great benefit in selected patients but are poorly utilised. Fewer than 5% of patients with heart failure receive heart failure pacemakers and many fewer receive heart pumps or transplantation. Earlier studies have shown that, a major reason is that heart failure patients are generally cared for by generalist doctors with limited awareness of these treatments.

The findings of the Karolinska research suggested that many more people need these treatments and should be referred to heart failure specialists for evaluation. The study was financed by the Swedish Heart-Lung Foundation, the Stockholm County Council and the Swedish Association of Local Authorities and Regions.

A study undertaken by Oxford and Bristol university researchers, reported by Pigott (2015), showed that children in the ‘general population’ performed better at GCSE examinations at age 16 than did vulnerable children in foster care, who in turn performed better than vulnerable children who remain with troubled families.

The GCSE results of 640,000 teenagers in England in 2013 were examined. Around 14,000 of these were deemed to be 'in need' but were still living with their birth parents, supported by social workers, and their results were not as good as the total of just over 6,000 pupils who were in care. At GCSE, 16-year-olds who had spent at least a year in foster care achieved better grades, equivalent to at least six grades overall, than children in other forms of care.

The theory is that children did better at school 'once they felt safe and secure'.

Alison Cook and Christy Glass (2014) examined the glass cliff theory, which predicts that occupational minorities are more likely to be promoted to leadership positions in organisations that are struggling, in crisis, or at risk to fail.

Between 1996 and 2010, in total, there were 551 transitions where a traditional (white male) leader was appointed CEO, whereas there were only 57 transitions where an occupational minority was appointed CEO. ‘There were 28 transitions where a traditional leader followed an occupational minority and only 4 transitions where an occupational minority leader followed an occupational minority. (Cook and Glass, 2014, p. 1083)’

To test the hypotheses, the authors ‘constructed a dataset of all CEO transitions within the Fortune 500 from 1996 to were collected using several reference websites such as,,,,, along with company websites….’

Our dependent variable to test the glass cliff theory is the transition of an occupational minority to CEO. An occupational minority appointed CEO was coded as a one, and a white man appointed CEO was coded as a zero. Our dependent variable to test the savior effect is the transition of a traditional white male leader replacing an occupational minority CEO. This transition was compared to the transition of a white man replacing a white man CEO and the transition of an occupational minority replacing an occupational minority CEO. The results showed, for example, that a white male CEO leader replacing an occupational minority CEO is significantly and negatively correlated to a firm’s return on equity  (Cook and Glass, 2014, p 1083).

Consistent with the glass cliff theory, the results of the statistical analysis showed that occupational minorities were ‘more likely than white men to be promoted CEO in firms experiencing short-, medium-, or long-term declines’. Furthermore, ‘negative firm performance in the short, medium, or longer term leads to the replacement of occupational minority CEOs with white men, a process we term the savior effect’. However, contrary to the author’s expectation there was no evidence that ‘occupational minorities experience shorter average tenures compared to white men’ (Cook and Glass, 2014, p. 1087).

Lorine Hughes et al. (2015) undertook a cross-national study of the link between economic dominance, indicators of culture, and homicide. Measures of social structure were derived from the World Bank’s World Development Indicators database. Measures of culture were from Waves 3 and 4 of the World Values Survey, a cross-national data collection effort begun by the European Values Survey group in 1981 and expanded to over 100,000 respondents representing 70 countries in wave four (1999–2004). Serious crime measurement was based on the average homicide rate reported in the 5th through 9th United Nations Survey of Crime Trends and Operations of Criminal Justice Systems, a periodic survey of government agencies conducted under the auspices of the United Nations Centre for International Crime Prevention (Office of Drug Control and Crime Prevention, Vienna).

Results from regression models suggest, inter alia, that homicide occurs most often in countries where free-market principles and practices drive the economy and where core cultural commitments are oriented toward achievement, individualism, fetishism of money, and universalism.

Another example used databases from two countries. Markus Gangl (2004) examined whether, in both the United States and in Western Germany, workers’ search for adequate reemployment after a period of unemployment was aided by welfare support. Are workers’ risks of incurring severe earnings losses, of experiencing occupational mobility, and of entering unstable job arrangements reduced by welfare support during unemployment?

Gangl used the Survey of Income and Program Participation (SIPP; U.S. Bureau of the Census 1984, 1991) and the German Socio-Economic Panel (GSOEP; see DIW 1999; SOEP Group 2001). These are both household panel surveys representative of each country's residential population. The surveys provide data employment, labour markets and job dynamics as well as social and institutional background information.

The analysis combined data from the SIPP Panels 1984, 1986, 1988, 1990, 1992, and 1993 and the West German data from GSOEP waves A-M (samples A+B) to generate monthly employment history information for the 12-year period January 1984 to December 1995. The data focused on unemployment spells among mid-career workers who were in employment immediately preceding the unemployment spell, for whom unemployment is strongly associated with job loss, and for whom unemployment is thus likely to come as an interruption of careers.

The sample deliberately excluded both unemployment spells of entrants to the labour force as well as job search periods of (predominantly) women returning to the labour market after career interruptions related to, for example, child or elderly care. In doing so, it is intended to focus on the role of the welfare state in mitigating the consequences of job loss among the core male and female workforce highly attached to the labour market.

The combined SIPP data yielded a sample of 24,100 unemployment spells of 21,551 workers who were observed for a total of 98,749 months in unemployment. The smaller GSOEP database generated a total of 3,251 unemployment spells of 2,264 workers who were observed for a total of 32,498 months of unemployment.

In addition to the core spell information, the databases included gender, age, ethnicity, workers’ education (including completion of vocational training in the German sample), labour force experience, tenure, occupation, industry, and earnings with the previous employer as main worker-level characteristics; they also included quarterly vacancy ratio as a measure of aggregate labour market dynamics.

To test the prediction that unemployment insurance enables workers to avoid more severe career breaks, each outcome dimension has been measured against six job quality measures in total: (1) the incidence of a real earnings loss compared to the worker's pre-unemployment job, (2) the incidence of an earnings loss of at least 20% of the worker's pre-unemployment real earnings, (3) the incidence of occupational mobility across two-digit SOC80, respectively two-digit ISCO-68 occupations, (4) the incidence of downward status mobility as measured by the ISEI occupational status scale [see (accessed 29 June 2015) for discussion of international measures], (5) entry into a new job that lasts less than 12 months, and (6) entry into a job that lasts less than six months.

If search theory is correct about unemployment insurance enabling workers to avoid more severe losses, positive effects of unemployment insurance should be more pronounced for the more severe measures (2), (4), and (6). Given that search theory focuses on welfare states reducing workers’ career risks, the analyses will not consider the likelihood of upward mobility, significant earnings gains, or similar positive returns to unemployment.

The subsequent statistical analysis was fairly complex and based on a two-equation model that describes unemployment processes from the rate at which workers exit unemployment and go back into employment and the job quality workers attain upon re-entering employment.

The study concluded:

For both U.S. and German workers, the analysis provides evidence that unemployment insurance improves the quality of post-unemployment jobs. When covered by benefits, workers were found to be better able to avoid earnings losses, occupational mobility, and job instability than they were without transfer income. Consistent with the basic job search model, unemployment insurance primarily permits workers to prevent cumulative disadvantages that may otherwise have followed from the financial pressure on workers to find reemployment quickly. The empirical analysis is clear about the fact that unemployed workers face constrained choices, so that reemployment rates and quality of reemployment are negatively related: if workers want to escape unemployment quickly, they will typically have to compromise on job features by accepting lower earnings, lower levels of job stability, and jobs in occupations other than the one in which they have previously been working.


Next 7.6.2.Phenomenology and secondary analysis