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 2012–2020

Page updated 29 April, 2020

Citation reference: Harvey, L., 2012–2020, 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

7.2.1 Minimal re-analysis
7.2.2 Significant re-analysis
7.2.3 Meta-analysis

7.3 Nature of the data
7.4 Data sources

7.5 Examining data sources
7.6 Methodological approaches
7.7 Summary and conclusion

7.2 Extent of re-analysis of secondary data
The extent of secondary analysis varies considerably from minimal reworking to extensive reanalysis.

7.2.1 Minimal re-analysis
Analysis may be, for example, identifying and drawing attention to extant data, statistics or examples and relating them to a hypothesis, thesis or policy issue. A reported increase in death rate from alcohol consumption is such an example (see CASE STUDY Alcohol deaths). This example takes the data from the Office of National Statistics and shows the increase in rates of death between 1991 and 2004 and also breaks this down by gender. The analysis is linked to policy initiatives to try and reverse the trend.

Biggs (2015, p. 141) showed that the literature on political participation asserts that protest has increased in the four decades from 1970. He reexamined the literature and showed that the claim is misleading. The trend in the literature identified by Biggs, is derived from surveys 'asking questions about participation in various forms of protest, including demonstrations, boycotts, and unofficial strikes'. The problem with this is that such a definition of protest was relevant in the United Kingdom in the 1970s but is less so in the 21st Century. The failure to take into account official strikes distorts the perception of protest as official strikes in the 1980s and 1990s 'greatly outnumbered demonstrations and other forms of protest'. The decline in the number of strikes since the 1980s 'more than offsets any increase in demonstrations and boycotts, meaning that the total volume of protest has decreased'. This shows that one has to be wary of drawing conclusions without a careful checking of the data sources.

Similarly, Biles and Dalton (1999) showed that, overall, the death rates in Australian public and the newly-created private prisons were very similar. They used offical Australian data and calculated the number and rate of deaths and suicides in both public and private prisons over the period 1990 to 1999. They concluded that the rate of death from all causes in private prisons is 2.93 per 1000 prisoners per year (95% confidence interval 1.99–3.88) compared with 3.35 per 1000 prisoners per year (95% confidence interval 3.04–3.66) in public prisons. The difference is not statistically significant (as the two confidence intervals overlap).

They go on to say:

The difference between private and public prisons in relation to suicide is apparently not so great, however, with the rate for private prisons being 1.51 per 1000 prisoners per year (CI 0.83–2.18), compared with 1.57 per 1000 prisoners per year (CI 1.36–1.78) for public prisons. These two rates are clearly very close and their difference not significant.

Another example, by Curry et al. (2004) used a sample of old data from 1991 on offenders who had been convicted of three violent crimes in Texas. They looked at whether offender and victim gender in violent crimes impacted on sentence length. As they state:

Data came from a project entitled Sentencing Dynamics Study: A Sourcebook of Felony Sentencing Practices in Urban Texas in 1991, a data collection effort mandated by the 72nd Texas Legislature. A random sample consisted of 7,729 offenders convicted of a felony between January 1 and September 30, 1991, in what were then the seven largest Texas metropolitan counties (i.e., Bexar, Dallas, El Paso, Harris, Nueces, Tarrant, and Travis) for 10 major categories of crime, representing 93.3% of all convictions in these counties (Fabelo, 1993). Offenders were randomly sampled within each offense category and each county. Because the gender of crime victims is the primary concern of this research, we restricted analyses to those 1,242 offenders convicted of assault, robbery, and homicide. An advantage of these data was that they represented seven large metro counties rather than a single court or jurisdiction in which idiosyncratic effects could alter results. A second advantage was that, because they came from the same state, there were no jurisdictional differences in terms of legal definitions and proscribed punishments (cf., Spohn & Beichner, 2000).

They found that offenders who victimised females received substantially longer sentences than offenders who victimised males. Results also showed that victim gender effects on sentence length are conditioned by offender gender, such that male offenders who victimise females received the longest sentence of any other victim gender/offender gender combination. However, whereas these effects are observed for sentence length, no victim gender effects are observed on whether offenders received an incarcerative or non-incarcerative sentence.


Next 7.2.2 Significant re-analysis