Your results chapter says, “The programme reduced recidivism from 24% to 18%.” Your supervisor asks how you know the programme made the difference. Start by opening both sources. For each figure, record what counted as an event, when follow-up began, how long it lasted and how many people had a known outcome. Check who those people were. If 18% measures police contact among programme completers and 24% measures reconviction in a different group over a longer period, report them as separate findings. You can compare the figures directly once you have established that the available data let you calculate the same outcome for both groups.
In a thesis in the “Resocialisation” programme, a statement about a difference between two groups and a claim about a programme's effect require different evidence. The percentages may be correct while the programme's role in the difference remains unknown. The fictional example below shows how to revise the results, explain what data were available and draw a conclusion that reflects what you learned. It focuses on editing a thesis and interpreting research.
Establish what counted as recidivism
In social rehabilitation research, “recidivism” can name a general research problem or a specific measure. If it refers to a measure, the methods section should define it. Check whether the source counts reconviction, readmission to prison, recorded police contact in connection with a suspected new offence or another event. Each captures something different. A person can appear in one register without having an event recorded in another. Name the event as the researchers defined it and identify the data source so readers know who was counted.
When should follow-up start?
Follow-up can start on release from an institution, on completion of a sentence, at the start of a programme or on completion of it. The start date determines which events count. Suppose you follow one group for a year after release and another for a year after completing a three-month programme that began after release. The dates covered differ even though both follow-up periods last a year. Record the event that starts follow-up and check whether the window is calculated for each person or uses the same calendar dates for everyone.
Distinguish the date recorded in the register from the date of the underlying behaviour. Suppose a person leaves an institution on 10 January 2024. In this exercise, the records give 20 November 2024 as the date of a new offence and 20 April 2025 as the date the conviction became final. In a 12-month window following release and ending on 10 January 2025, the offence falls within the window and the final conviction comes later. A measure that counts offences that later led to a conviction could include this case if the outcome was defined that way and the data were checked late enough. A measure of convictions that became final within the year would exclude it. Apply the same timing rule to everyone.
Who belongs in the numerator and denominator?
Before calculating anything, identify the unit. If you want the percentage of people with at least one reconviction, the unit is the person. Three convictions within the same period add one person to the numerator. If you count all convictions, three convictions count as three events. Depending on the research question, you could report the number of events per hundred people or the distribution of the number of convictions within the group. Both descriptions can be useful in the chapter when the unit is clear.
Check which stage the denominator represents. The numbers of people referred to the programme, who started it, who completed it and whose one-year outcomes are available can differ. For each version of the measure, the answer to “Who does this percentage describe?” can change. If the report gives a percentage alone, look for the number of people and the exclusion rules. If you cannot establish the sample size, write “the report gives a percentage” and state what information you need to make the comparison.
How do you distinguish two percentages in the same table?
One row might show the percentage of a particular group with at least one conviction. Another might describe the groups represented among all people with a reconviction. If 12 out of 60 people in the group experienced the event, the percentage within that group is 20%. If those 12 people also account for 12 out of 30 reconvicted people across the study, 40% describes their share of the people with the event. The numerator is the same, but each denominator answers a different question. Read the heading, caption and any note beneath the table before using a figure. The counts reported by the source may be the quickest way to check your wording.
Polish sources illustrate different ways of counting
Published reports can help you see how to describe each measure precisely. Start with the methods and the notes below the table. Record the report year or period and the table number alongside the percentage. This helps you find the figure again when you move from the methods to the discussion chapter. The sources below concern particular datasets and groups and illustrate claims you can check against the original.
Ministry of Justice report – one title, several measures
The Ministry of Justice published Powrotność do przestępstwa w latach 2009-2015 in 2017, drawing on Poland's National Criminal Register (KRK). Chapter II describes five-year periods after a conviction became final. Chapter III describes five-year periods after release from a prison or remand centre. The opening chapter separately reports annual counts of convictions under the Polish legal category of special recidivism. Check which chapter a figure comes from before citing it. If you group the figures under one label, give the definitions so readers can see which events and people each part covers.
If your chapter concerns the first year after release, choose the part of the report that starts follow-up at release. Check whether your question concerns the proportion of the whole group or the distribution of time to another offence among people who reoffended. This matters when percentages appear side by side in a chart. You might write, “The report covers both the period after a conviction became final and the period after release.” If you give a specific proportion, identify the denominator and cite the relevant passage.
IWS study – a shared window and differences that predate follow-up
In a 2024 report for the Polish Institute of Justice, Paweł Ostaszewski, Joanna Klimczak and Justyna Włodarczyk-Madejska analysed reconvictions after people completed a custodial sentence. They compared a sample who served their sentences under electronic monitoring (SDE) with a prison sample, using a three-year window from the end of the sentence. They explained that the criminal records let them identify conviction dates. Exact dates for every offence were unavailable. The methods subsection and the section introducing the results explain which cases they could identify and how they selected the follow-up window. Keep their definition when citing a result.
When reading the comparison, note that the researchers drew random samples from two existing databases after the way each person's sentence would be served had already been decided. Random sampling within groups and random assignment to study conditions are different procedures. In the modelling section, the authors considered prior convictions and how people were selected for electronic monitoring, among other factors. They explained that the available data could not establish whether the difference arose from the way the sentence was served or from other factors. In your thesis, use this example to explain why you need to describe how groups were formed even when they share a follow-up window.
Drapała and Kulma – one sample, successive follow-up windows
Drapała and Kulma (2014) studied 243 people who had left juvenile correctional institutions. In their analysis of time to reconviction, they reported that approximately 18% of the sample had been reconvicted by one year after release, 44% by two years and 58% by three years. The windows in the chart are cumulative. A person counted in the first year is also counted by the end of the second and third years. This example shows why a comparison of one-year and three-year follow-up requires restricting both sets of data to a common period. The three percentages describe the sample and the counting rule in that publication.
You can use this example if your supervisor comments on different follow-up lengths. If you write “the percentage increases as the window gets longer”, explain that you mean the cumulative proportion with at least one event under a specified definition. If the source reports separate percentages for each year, revise the description to reflect that reporting method. In either case, the relationship between shorter and longer follow-up depends on whether the population in the denominator changed and who could remain under observation.
Revise the first sentence when the sources use different definitions
Suppose you are writing about a voluntary post-release support programme called “Back to Everyday Life”. The programme and all the counts below are fictional. Participants could start attending sessions after release. Of the 120 people who started, 100 completed the programme. The student has a short summary about those who completed the sessions and a report about people from a different institution who did not take part in the programme. In the first draft, the student wrote, “The programme reduced recidivism from 24% to 18%.” Before revising it, the student works out what each number represents.
What does 18% actually mean in the programme summary?
The summary identifies 18 out of 100 people who completed the sessions and had at least one recorded police contact in connection with a suspected new offence within 12 months after completion. Dividing 18 / 100 gives 18%. People who started the programme and did not complete it are outside the denominator. The event is a recorded contact in specific circumstances. The summary gives no information on whether a conviction subsequently became final or how many people had a final conviction within the same period.
You could write in the results, “Of the 100 people who completed the programme, 18 had at least one recorded police contact in connection with a suspected new offence within 12 months after completion (18%).” The first part defines who was counted. The rest gives the event and the window. If you have a category for everyone who started the programme, you can describe their outcomes separately if the document provides them. In this summary, the 20 people who did not complete the programme have no outcome reported in this way.
What does 24% mean in the other institution's report?
In the second source, 24 out of 100 people who left a different institution had at least one new conviction that became final within two years of release. The percentage is 24 / 100 = 24%. A descriptive sentence would read, “The report from the second institution recorded at least one new conviction that became final within 24 months of release for 24 of the 100 people who left the institution (24%).” In your own thesis, cite the source and relevant pages for each percentage.
You can put the two figures side by side if you need to illustrate what the sources cover. The numerical difference of six percentage points between 18 and 24 does not measure a difference for a shared outcome. The event, start date and follow-up period all differ. The first source covers only programme completers, and the second covers people leaving a different institution. To revise the paragraph, follow the two figures with a brief explanation. “The sources used different events and follow-up periods. Each percentage describes the group according to the definitions in its report.”
How do you describe this problem if you only have summary reports?
Changing “reduced” to “was associated with” leaves the difference of six percentage points looking as if it came from a comparison based on one definition. Check whether the source with the two-year result also reports a one-year result for the same event and population. Then check whether the other source reports an event matching that definition. If you cannot obtain a common outcome, report the studies in separate sentences, explain the limitation and discuss them together in terms of methods. For example, “The reports illustrate how studies of recidivism can use different events and follow-up start dates.”
Dividing the 24-month percentage by two to get a one-year figure assumes a pattern of events over time that the data do not establish. If you have the total number of events rather than the number of people with a first event, dividing by the number of participants cannot tell you how many people had an event. You need to know whether the events were recorded for different people or whether some people had more than one. If your study gives you access to the detailed records, you can calculate the outcome using the same start date, window and definition in both groups. The next part of the exercise assumes access to such an additional fictional dataset.
Recalculate the outcome on a common basis if the data allow it
In the second part of the exercise, the student receives a separate fictional dataset. It allows the student to identify new convictions that became final within 12 months after release, using the same rules for both groups. The start event was chosen because it can also be identified for people who did not take part in the programme. Everyone in both groups left their institution in the same calendar year. The comparison includes 120 people who started “Back to Everyday Life” and 100 people from the other institution who did not take part. At this stage, 120 covers all programme starters, including those who later withdrew. In a real study, document the programme start dates and the reasons for choosing those groups. These details also help explain differences in how the groups were formed.
How do you obtain comparable information for both groups?
- Use the same event. Look for at least one new conviction that became final. Use the date on which the conviction became final as the event date. If two convictions for one person became final within the window, that person still counts once in the numerator.
- Start the window at the same event. Each person's follow-up begins on their release date. Count events after that date up to and including the first anniversary. For the other institution, use the first year of its available two-year follow-up. A person with no event at the end of the year can be classified as having had no event only if their status during that window can be established reliably.
- Distinguish the outcome from whether it is known. In the exercise, it was possible to establish whether the event occurred for 112 of the 120 people who started the programme and all 100 people in the comparison group. The information needed to determine the outcome could not be linked for eight programme starters. Their outcomes remain unknown.
- Count people with known outcomes. Of the 112 programme starters with known outcomes, 14 had the event and 98 did not. In the second group, 16 had the event and 84 did not during the year. In line with the earlier report from the other institution, 24 people in total had the event within two years. For eight people, the first qualifying event fell after the first anniversary of release, up to and including the second anniversary.
- Report each percentage with the correct denominator. Among people with known outcomes, the percentages are 14 / 112 = 12.5% and 16 / 100 = 16.0%. In these records, the percentage for programme starters was lower by 3.5 percentage points. This is a descriptive comparison within the specified follow-up window. People with unknown outcomes account for
8 / 120, or approximately 6.7% of all programme starters.
Check why 12.5% can stand alongside the earlier 18% even though both figures concern people associated with the programme. The first report counted police contacts after the sessions ended among 100 programme completers. The second calculation counts convictions that became final after release among programme starters whose outcomes were known. The figures use different definitions and cover different people, so one cannot be used to verify the other. In your own thesis, report them together only if you address both questions and can explain what each number means.
How do the eight unknown outcomes affect the comparison?
Dividing 14 by 120 assumes that none of the eight people with unknown outcomes had the event. If the study could not confirm this, report the value as the lower bound in a simple sensitivity check. In this example, you can calculate two bounds for all 120 people. If none of the eight people had the event, the proportion is 14 / 120, or about 11.7%. If all eight had the event, it is (14 + 8) / 120, or about 18.3%. This range shows what is possible under two extreme assumptions about the missing data. It has no interpretation as a confidence interval.
The comparison group's 16.0% lies between those values. Some possible outcomes for the eight people would produce a lower percentage for programme starters, and others a higher one. With these denominators, the percentage would exceed 16.0% if at least six of the eight people with unknown outcomes had the event. Five events would give 19 / 120, or approximately 15.8%, and six would give 20 / 120, or about 16.7%. These are calculations under hypothetical scenarios. The available data give no basis for saying which scenario is likely.
In a real dataset, investigate why the outcomes are missing. The cause could be a record linkage error, a change in where follow-up took place or the fact that some people have yet to be observed for a full year. The appropriate response can differ. If you know that a person had the event in month nine, you know they met the criterion within the year. If no event was recorded by month nine and you have no later data, the annual outcome remains unknown. This distinguishes a confirmed event from incomplete follow-up. If many people have different lengths of follow-up, address that when choosing how to analyse the data.
When should you use “percentage points” and when should you report a relative difference?
Subtracting 12.5% from 16.0% gives 3.5 percentage points. This wording makes the difference between percentages on a common scale easy to read. The ratio 12.5 / 16.0 is about 0.78. The observed proportion among programme starters with known outcomes was therefore approximately 78% of the proportion in the comparison group. A relative difference of about 22% is meaningful only if you specify the reference group and compare two outcomes measured in the same way. “The programme reduced risk by 22%” goes beyond what this calculation supports. For this example, give both percentages and the descriptive difference. This leads to the question of how the groups were formed.
Check what a common definition still leaves unresolved
Once the windows are aligned, you can describe a difference based on the same definition. To attribute the difference to the programme, first establish whether the groups give you a credible basis for judging what would have happened to participants in comparable circumstances without the programme. In the exercise, people from one institution volunteered for the programme. The comparison group came from another institution. With this design, the effect of programme participation cannot be separated from the wider circumstances of the two institutions. Applying the same rules to the register makes the events easier to compare. Any earlier differences between the groups still matter.
How can the way participants were selected affect the percentages?
People who volunteer may differ from people who never had the opportunity to enrol. They could have different experiences with the justice system, access to support after release or willingness to seek help. Look for information on these points in the methods before using them as possible explanations. “The groups may have differed in access to support” describes a possibility. “The groups differed in access to support” requires evidence showing how it was measured and what the study found. Age alone can tell you little about differences in the groups' histories. If the source describes how people were selected, report which criteria could be verified for both groups and which remain unknown.
Check whether you measure programme participation from the invitation, the first session or completion of the last session. To enter a group defined by completing several weeks or months of sessions, participants must stay until the end. People who withdrew or could not continue fall outside the group. In the exercise, 20 of the 120 programme starters did not finish. If a report covers only completers, limit your description to that group. The revised comparison uses programme starters so that completion is no longer required for inclusion in the denominator. You still need to know when people actually started and what happened between release and their first session.
What if some people started sessions after follow-up had begun?
A shared start event is practical, but it can precede actual participation. A person might leave the institution in January and attend their first session in March. Part of the year of follow-up then occurs before they started. You can say that the groups were defined by whether people later started the programme during follow-up. To evaluate effectiveness, the study design should define when participants start receiving the intervention and use comparable times to establish eligibility. The time between release and the first session may be related to whether people can enter the programme group. Avoid attributing events from that period to an intervention that had yet to begin.
If you have the data, record the date of release, the first session, completion of the programme and the event separately so you can check their order. In a methods paper on analysing observational data, Fu and colleagues (2026) show why the timing of eligibility, group assignment and the start of follow-up needs to align in studies that estimate intervention effects. Apply this principle when reading your study plan. Check what was known about participation when follow-up began. If you only know whether people took part later, describe the comparison as one between groups defined by that later participation. If your supervisor proposes new dates for the window, record which people and events the analysis will include and when you made the decision.
Can a more complex analysis resolve the problem?
You may come across analyses that compare people with similar characteristics or use a model to account for observed differences. Record which characteristics were measured before the programme and which were included in the analysis. A model may change the estimated association after accounting for those characteristics. The conclusion still depends on how well the study identifies and accounts for factors related to participation and the outcome, and on the quality of measurement. If the report gives an adjusted estimate, present it separately from the raw percentages and name the measure. A difference in percentage points, a risk ratio and an odds ratio mean different things.
If you are reviewing your own chapter, look closely at phrases like “effectiveness was confirmed” and “after accounting for differences”. The second requires you to name the variables or assumptions involved. The first requires a study design capable of addressing the question of effect. The IWS example shows that authors can report group comparisons and multivariable analyses while also explaining that they cannot clearly separate the role of how the sentence was served from other factors. Describe what your own evidence supports.
Revise the results, discussion and conclusion to reflect the data
After correcting the calculations, reread every place where the original claim appears. A causal claim can remain in the abstract, figure caption or conclusion even when the tables and methods use the right terms. In the example, the description depends on whether you have only summaries that use different definitions, or also have data to calculate a common one-year measure. In the first case, report 18% and 24% with their definitions and groups. In the second, report the percentages based on the common measure and explain how the selection of participants and the missing data limit the conclusion.
Results based on the same event definition
Within 12 months after release, a new conviction became final for 14 of the 112 programme starters with known outcomes (12.5%) and for 16 of the 100 people from the other institution (16.0%). Among people with known outcomes, the percentage for programme starters was 3.5 percentage points lower. Outcomes could not be established for eight of the 120 people who started the programme.
In a real study, give the source name and the date you accessed the data when reporting the results. In the methods, explain how you determined the outcome and follow-up start date. If the numbers of people differ between paragraphs, explain why. “The analysis included 220 people” leaves open how many had a confirmed one-year outcome. In this example, 220 people were included in the description of the groups and 212 had known outcomes that could be used in the comparison.
Discussion of how much the comparison can tell you
The comparison used one event definition and the same follow-up window after release. The groups came from different institutions and programme participation was voluntary. The observed difference may reflect how participants were selected and differences in circumstances after release. One-year outcomes could not be established for eight programme starters. Under extreme assumptions about their outcomes, the percentage for this entire group could range from 11.7% to 18.3%. This range includes the 16.0% observed in the comparison group.
Notice how the paragraph moves from the figure to its limitations. First state how the comparison was made. Then explain what affects whether the difference can be attributed to the programme. Show where missing data can change even the direction of the comparison. You can draw on a source describing a comparable event and group in your own discussion. If it uses a different window, explain that when citing it. Studies of other populations can provide context when you keep their scope clear.
A conclusion that fits the study design
In records with known outcomes, the percentage with a new conviction that became final within a year was lower among programme starters than among people in the comparison group from the other institution. The study design and the unknown outcomes for some people prevent attributing the difference to programme participation.
If your supervisor wants a single sentence answering the research question, shorten the description while retaining its limits. “The percentage was lower in the records analysed, and the programme's effect remains uncertain” lets readers understand what sort of finding you are reporting. Uncertainty alone provides no evidence that the programme was ineffective. The result may justify collecting data for a further study if you specify what information is needed, such as outcomes for people without complete follow-up or the criteria for joining the two groups.
What if the research question still promises to assess effectiveness?
Before submitting the thesis, compare the wording of your aim, research question and conclusion. If all three concern the “effect of participation” and your data can only describe two percentages, discuss narrowing the question with your supervisor. You could ask, “What percentages of people in the two observed groups had a new conviction that became final within 12 months of release?” The available data can answer that question and it retains a defined comparison. If you have only two reports with different definitions, ask, “How do the two reports measure recidivism, and what can we learn by comparing them?” If you are also working with interviews and institutional records, see the guide to social rehabilitation research, documents and conclusions. It can help you relate each finding to the kind of evidence you have.
Review the draft before sending it back to your supervisor
Start with subheadings and captions. Give a chart showing 12.5% and 16.0% a title that matches its content, such as “Reconvictions that became final within a year of release among people with known outcomes”. Give the sample sizes in the caption and explain missing data in the text. If the chart used the two earlier, incompatible percentages, first decide whether you still need it after revising your argument. A chart comparing two percentages can suggest that they share a definition even if a footnote explains the differences. Readers should be able to tell what the figures mean from the labels.
What should you check in every sentence with a figure?
- Event: Does the type of record match the term used in the sentence? Describe recorded police contact, conviction and admission to an institution using the study's definitions.
- Time: What starts follow-up, and which event date governs inclusion? For data from different years, check whether observations cover comparable circumstances.
- Group: Does the percentage describe people invited, people who started or completed the programme, or people with a known outcome? Write down the numerator and denominator, including for your own calculations.
- Missing data: How many people have unknown outcomes, and what do you know about why? If you show a range based on assumptions, distinguish it from the observed percentage.
- Type of claim: Are you reporting an observed difference, an association estimated in a model or a programme effect? Check what the evidence supports.
- Consistency across the thesis: Do the aim, methods, captions and abstract describe the same event, window and group? If an earlier definition still appears, attribute it to the appropriate source or use wording that matches the new calculation.
When sending the passage back to your supervisor, state how the comparison changed. Include the original definitions and say whether you have the data needed to calculate a common measure. If the analysis is complete, report how many people had unknown outcomes and show how you obtained the figures used in the chapter. This helps you check whether the revision covers the conclusion and the calculation behind it.
Questions about conclusions in a social rehabilitation thesis
How does thesis scope differ from study limitations?
Scope defines the deliberately chosen problem, evidence, population and period. Limitations explain how design features or difficulties restrict interpretation. An omitted topic is not automatically a flaw; the important question is whether the conclusions stay within the stated scope.
How can I connect a limitation to its effect on the conclusions?
Identify a specific feature of the procedure, the finding or claim it affects, and the consequence. Then explain what you actually did to address it and what remains uncertain. A generic list mentioning sample size and time pressure is not enough.
Are the opinions of educators sufficient to evaluate the effectiveness of a resocialisation programme?
The opinions of educators allow you to present their perspective, experiences, and assessment of the activities. They are not sufficient on their own to demonstrate the effectiveness of the programme. I will help separate the accounts of the participants from conclusions about the effects and tailor the scope of interpretation to the research design and available data.
When analysing survey results, will you account for missing responses and questions that allow multiple answers?
Yes. Send the questionnaire and the spreadsheet with responses, along with an explanation of the notation used. I will help organise the counts, the base for calculating percentages, and table descriptions, so it is clear how many people the summary covers and how missing responses were handled.
What should I send if I already have feedback from my supervisor?
Send the current version of the thesis, the supervisor's comments, guidelines, and the deadline. If the comments concern a specific chapter or results, highlight those sections. Based on this, I will define the scope of changes, the price, and the possible completion date.
What should I send for a quote?
Provide your topic, subject area, type of thesis, stage of preparation and required deadline. Describe what needs to be prepared, checked or revised. Attach the current document, university guidelines and any comments from your supervisor.
For research support, include your research outline, survey or interview guide and the data you have collected. For editing that addresses text similarity, attach your JSA report if you have one. You don't need every document before asking about the scope of support.
Sources
- Ministerstwo Sprawiedliwości, Departament Strategii i Funduszy Europejskich, Wydział Statystycznej Informacji Zarządczej (2017). Powrotność do przestępstwa w latach 2009-2015.
https://isws.ms.gov.pl/pl/baza-statystyczna/publikacje/download%2C3502%2C14.html - Ostaszewski, P., Klimczak, J., Włodarczyk-Madejska, J. (2024). Powrotność do przestępstwa skazanych, którzy odbyli karę pozbawienia wolności w Systemie Dozoru Elektronicznego (SDE). Instytut Wymiaru Sprawiedliwości.
https://iws.gov.pl/wp-content/uploads/PDF/2024/ANALIZA%20WYMIARU%20SPRAWIEDLIWO%C5%9ACI/2024_AWS_P_Ostaszewski_J_Klimczak_J_Wlodarczyk-Madejska_-Powrotnosc-do-przestepstwa-skazanych-ktorzy-odbyli-kare-pozbawienia-wolnosci.pdf - Drapała, K., Kulma, R. (2014). Powrotność do przestępstwa nieletnich opuszczających zakłady poprawcze (raport z badań). Prawo w Działaniu. Sprawy Karne, 19, 204-246.
https://iws.gov.pl/pwd-19/ - Fu, E. L., Harhay, M. O., Schneeweiss, S., Desai, R. J., Hernán, M. A. (2026). Starting right: aligning eligibility and treatment assignment at time zero when emulating a target trial. BMJ, 392, e084909. DOI
10.1136/bmj-2025-084909.
Would you like help revising a conclusion about recidivism?
Share your thesis topic, your supervisor's comments and your draft description of the data and conclusions. Include anonymised counts, the event definition and the follow-up rules. Indicate where the chapter reports percentages so we can check that the methods, results and conclusion agree.
Contact StudentSupport


