What a PGSI Score Can and Cannot Tell You About Gambling Harm

Many Australians first meet the Problem Gambling Severity Index through a short online quiz. There are nine questions about the past year, four possible answers to each, and a final number. That number arrives with a label such as "low risk", "moderate risk" or "high risk", and many people read it as a diagnosis. It isn't one.

The Problem Gambling Severity Index, usually shortened to PGSI, is the most widely used gambling measure in the world. Prevalence surveys use it, help services use it, government reports use it, and so do many self-tests. It was built for one specific purpose, measuring the severity of gambling problems across a population, and that purpose shapes what the score means. It explains why two people with the same score can be in very different situations, why national statistics based on it often seem to contradict each other, and why a clinician uses different tools to diagnose gambling disorder.

This article explains how the PGSI works, what its categories mean, how it differs from a clinical diagnosis, and why the same index can produce such different numbers in different reports.

Where the PGSI comes from

The PGSI was published in 2001 by Jackie Ferris and Harold Wynne as part of the Canadian Problem Gambling Index, a larger questionnaire developed for the Canadian Centre on Substance Abuse. The full Canadian index included questions about gambling involvement and other correlates. The nine-item severity section was the part that spread. It became the standard measure in population gambling surveys in Australia, the United Kingdom, Canada and many other countries.

The PGSI was not designed as a clinical instrument. Its developers wanted a tool for general-population surveys that reflected a public health view of gambling, in which problems sit on a continuum rather than falling neatly into "has a disorder" or "does not". That design choice affects everything about how the score should be read.

The nine questions and how they are scored

Each question refers to the previous 12 months and asks how often something has happened. The four answers are scored as follows: never (0), sometimes (1), most of the time (2) and almost always (3). The total runs from 0 to 27.

The nine items cover these areas:

Betting beyond means: betting more than you could really afford to lose.
Tolerance: needing to gamble with larger amounts to get the same excitement.
Chasing losses: going back another day to try to win back money you lost.
Funding gambling: borrowing money or selling something to get money to gamble.
Self-perception: feeling that you might have a problem with gambling.
Health effects: gambling causing health problems, including stress or anxiety.
Social reaction: people criticising your betting or telling you that you have a problem, whether or not you thought it was true.
Financial effects: gambling causing financial problems for you or your household.
Guilt: feeling guilty about the way you gamble or what happens when you gamble.

The items mix two kinds of content. Some describe behaviour, such as chasing losses, increasing stakes and borrowing. Others describe consequences and perceptions: health effects, financial problems, other people's criticism and guilt. A high score can come from different combinations of the two, and that is one reason the same score can describe quite different situations.

The categories and the two sets of cut-offs

The original scoring guidance grouped totals into four categories. Since then, researchers have proposed revised boundaries, and both versions are still in use.

html

<table>
  <thead>
    <tr>
      <th>Category</th>
      <th>Original cut-offs (Ferris and Wynne, 2001)</th>
      <th>Revised cut-offs (Currie and colleagues)</th>
      <th>Labels often used in recent Australian reports</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>No reported problems</td>
      <td>0</td>
      <td>0</td>
      <td>Non-problem or no risk</td>
    </tr>
    <tr>
      <td>Low risk</td>
      <td>1 to 2</td>
      <td>1 to 4</td>
      <td>Low risk</td>
    </tr>
    <tr>
      <td>Moderate risk</td>
      <td>3 to 7</td>
      <td>5 to 7</td>
      <td>Moderate risk</td>
    </tr>
    <tr>
      <td>Highest category</td>
      <td>8 or more ("problem gambling")</td>
      <td>8 or more</td>
      <td>High risk</td>
    </tr>
  </tbody>
</table>

Three points matter here.

The top category is stable. Both systems treat a score of 8 or more as the highest category. Most of the disagreement concerns the lower bands.

The revised boundaries move people between categories. A person scoring 3 or 4 is "moderate risk" under the original scheme but "low risk" under the revised one. Two reports using the same data can therefore describe the same people differently. The revision came from work by Shawn Currie and colleagues, who found that the original lower boundaries did not separate groups with meaningfully different levels of harm as well as intended.

Labels have changed. Many Australian reports now call PGSI 8+ "high-risk gambling" rather than "problem gambling". The measurement is the same; only the name differs. Reading across older and newer reports means watching the cut-offs and the labels separately.

A PGSI score is not a diagnosis

A screening or severity measure and a diagnosis answer different questions. The PGSI asks how often certain experiences have happened over the past year. A diagnosis of gambling disorder asks whether a person meets defined clinical criteria, judged by a trained clinician who can take account of the wider context.

Two diagnostic systems are used internationally.

The American Psychiatric Association's Diagnostic and Statistical Manual of Mental Disorders, in its fifth edition (DSM-5) and the 2022 text revision (DSM-5-TR), defines gambling disorder as persistent and recurrent problematic gambling causing clinically significant impairment or distress. It requires at least four of nine criteria within a 12-month period. The criteria include needing to gamble with increasing amounts to achieve the desired excitement; restlessness or irritability when trying to cut down; repeated unsuccessful efforts to control or stop; preoccupation with gambling; gambling when distressed; chasing losses; lying to conceal the extent of gambling; jeopardising a significant relationship, job or opportunity; and relying on others for money to relieve a desperate financial situation. Severity is graded as mild (four or five criteria), moderate (six or seven) or severe (eight or nine). The behaviour must not be better explained by a manic episode.

DSM-5 made two notable changes from its predecessor. It renamed "pathological gambling" as "gambling disorder" and moved it into the same chapter as substance-related and addictive disorders, reflecting evidence of shared features. It also lowered the diagnostic threshold from five criteria to four and removed the criterion about committing illegal acts to finance gambling.

The World Health Organization's International Classification of Diseases, eleventh revision (ICD-11), includes gambling disorder under code 6C50, with separate codes for predominantly offline and predominantly online patterns. It describes a pattern of impaired control over gambling, increasing priority given to gambling over other interests and daily activities, and continuation or escalation despite negative consequences, with significant impairment in important areas of functioning. The pattern is normally evident over at least 12 months, although a diagnosis can be made sooner if all the requirements are met and the symptoms are severe.

html

<table>
  <thead>
    <tr>
      <th>Feature</th>
      <th>PGSI</th>
      <th>DSM-5-TR gambling disorder</th>
      <th>ICD-11 gambling disorder</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Main purpose</td>
      <td>Measuring severity of gambling problems in population surveys</td>
      <td>Clinical diagnosis</td>
      <td>Clinical diagnosis and health statistics</td>
    </tr>
    <tr>
      <td>Who completes it</td>
      <td>Usually self-reported</td>
      <td>Clinician assessment</td>
      <td>Clinician assessment</td>
    </tr>
    <tr>
      <td>Time frame</td>
      <td>Past 12 months</td>
      <td>12 months</td>
      <td>Usually at least 12 months</td>
    </tr>
    <tr>
      <td>Structure</td>
      <td>9 items scored 0 to 3, total 0 to 27</td>
      <td>9 criteria, at least 4 required</td>
      <td>Core features of impaired control, priority and continuation despite harm</td>
    </tr>
    <tr>
      <td>Output</td>
      <td>Score and risk category</td>
      <td>Diagnosis with mild, moderate or severe specifier</td>
      <td>Diagnosis, predominantly offline or online</td>
    </tr>
    <tr>
      <td>Considers other explanations</td>
      <td>No</td>
      <td>Yes, for example manic episodes</td>
      <td>Yes</td>
    </tr>
  </tbody>
</table>

The overlap between the systems is real but partial. Chasing losses, increasing stakes and borrowing money appear in both the PGSI and DSM-5. Other DSM-5 criteria, such as failed attempts to stop, irritability when cutting down, preoccupation and lying, have no direct PGSI equivalent. The PGSI includes items about health effects, criticism from others and guilt that are not diagnostic criteria at all. A person can score highly on the PGSI through consequences and guilt while showing few signs of impaired control, and the reverse is also possible.

How accurately do population screens identify disorder?

The most direct test of a screening tool is to compare its results with clinical assessment of the same people. In 2014, Robert Williams and Rachel Volberg published a study in International Gambling Studies comparing four instruments used in population research, including the Canadian Problem Gambling Index from which the PGSI comes, against clinical assessment. All four identified people without problems reasonably well. Their ability to identify people with problems was more limited. The Canadian index had weak positive predictive power at its standard threshold: a substantial share of people it classified as problem gamblers did not meet the clinical standard. Accuracy improved when the threshold was lowered to 5. The authors concluded that the overall classification accuracy of the most widely used instruments remained modest.

This pattern is common for any screening tool used in a population where the condition is relatively uncommon. Even a tool with good sensitivity and specificity produces a meaningful share of false positives when most people screened do not have the condition, simply because there are so many more people without it. For readers, the takeaway is that a high PGSI score is a strong signal worth taking seriously. It is not a confirmed clinical finding, and a clinician would need to look at the whole picture.

The reverse error matters too. A low score does not rule out serious harm. The PGSI depends on honest self-report, and people often understate gambling problems. Denial, shame and the tendency to compare oneself with heavier gamblers can all push scores down. Some harms, such as a damaged relationship or an emergency loan hidden from a partner, may sit only partly within what the nine items capture.

Why the low and moderate categories matter more than they seem

Because the top category gets most of the attention, it is easy to assume that harm is concentrated among people scoring 8 or more. Research suggests otherwise.

A 2016 study by Matthew Browne and colleagues for the Victorian Responsible Gambling Foundation estimated the total burden of gambling harm in Victoria, using methods similar to those used for diseases and injuries. People in the high-risk category experienced the most harm individually. But there are many more people in the low and moderate categories, and taken together they accounted for most of the population's harm. In the study's estimates, about half of the total harm burden came from low-risk gambling, about a third from moderate-risk gambling and about 15% from problem gambling. These figures are modelled estimates, built on assumptions about how harm translates into reduced quality of life, and they should be read as an order of magnitude rather than a precise measurement.

This finding is the main reason Australian public health agencies talk about gambling harm rather than only about problem gambling. A score of 2 or 4 is not a clean bill of health. It indicates that some negative experience is happening, even if it falls well short of a disorder.

Why gambling prevalence figures differ so much

PGSI-based statistics are reported widely in Australia, and they can appear inconsistent. Most of the differences come from a handful of methodological choices rather than contradictory findings.

Different denominators

The most common source of confusion is the base of the percentage. Some reports give the share of all adults, including people who did not gamble at all. Others give the share of people who gambled in the past 12 months. Since many adults do not gamble, the second figure is always higher.

The Gambling Commission's Gambling Survey for Great Britain shows the difference clearly. In its 2023 annual report, 2.5% of all adults scored 8 or more on the PGSI. Among adults who had gambled in the past 12 months, the figure was 4.2%. Both figures are correct; they describe different groups.

html

<table>
  <thead>
    <tr>
      <th>PGSI category (original cut-offs)</th>
      <th>Share of all adults</th>
      <th>Share of past-year gamblers</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Low risk (1 to 2)</td>
      <td>8.3%</td>
      <td>13.7%</td>
    </tr>
    <tr>
      <td>Moderate risk (3 to 7)</td>
      <td>3.7%</td>
      <td>6.0%</td>
    </tr>
    <tr>
      <td>Problem gambling (8 or more)</td>
      <td>2.5%</td>
      <td>4.2%</td>
    </tr>
  </tbody>
</table>

The figures come from the Gambling Survey for Great Britain annual report for 2023, published by the Gambling Commission. They are included here to show the effect of the denominator, not as Australian estimates.

Different surveys and samples

Australian data comes from several sources with different designs. A nationally representative survey by the Australian Gambling Research Centre at the Australian Institute of Family Studies, conducted in 2024 with 3,881 adults and published in September 2025, found that 65.1% of adults had gambled in the past year. It classified 7.6% of adults as low risk, 4.8% as moderate risk and 2.6% as high risk. That makes 15% of adults with some level of PGSI-defined risk, compared with 11% in a similar 2019 study.

The ANUpoll series by researchers at the Australian National University, using a different panel and a different measure of risky gambling, reported 19.4% in January 2025, up from 13.7% a year earlier. The two figures are not comparable. They differ in sample design, timing, categories and, in some cases, denominators.

Different modes and question formats

Whether a survey is conducted face to face, by phone, online or by post affects how people answer sensitive questions. People often report more stigmatised behaviour when answering privately online than when speaking to an interviewer. Small changes in question wording or answer options also affect results. Trends are most reliable when measured within a single survey series with a consistent method. Figures from different series should be compared with caution.

Different cut-offs

As shown earlier, the original and revised boundaries move people between the low and moderate categories. Reports that do not state which set they use cannot be compared directly at those levels.

The practical rule is to check four things before comparing any two gambling prevalence figures: the population, the denominator, the survey method and the cut-offs. If any of them differ, the comparison needs qualification.

Other screening tools you may meet

The PGSI is not the only short gambling screen. Clinics and general practice often use even briefer tools designed to flag people for further assessment.

The Lie/Bet questionnaire has two questions: whether a person has ever felt the need to bet more and more money, and whether they have ever had to lie to people important to them about how much they gambled. A "yes" to either suggests further assessment.

The Brief Biosocial Gambling Screen has three questions covering irritability when trying to cut down, hiding gambling from family or friends, and needing financial help because of gambling. It was developed to identify people likely to meet diagnostic criteria.

The NODS-CLiP is a three-item screen derived from a longer diagnostic interview, focusing on control, lying and preoccupation.

These tools are designed for speed rather than detailed measurement. Like the PGSI, a positive result is a reason for a proper conversation, not a diagnosis.

What the evidence supports and what it does not

html

<table>
  <thead>
    <tr>
      <th>Statement</th>
      <th>Assessment</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>A PGSI score of 8 or more means a person has gambling disorder</td>
      <td>Not supported. The score signals serious gambling problems and warrants assessment, but diagnosis requires clinical criteria and judgement.</td>
    </tr>
    <tr>
      <td>A low PGSI score means gambling is causing no harm</td>
      <td>Not supported. Low and moderate categories involve real negative experiences and account for a large share of total population harm.</td>
    </tr>
    <tr>
      <td>The PGSI is a reliable tool for comparing gambling problems across a population over time</td>
      <td>Broadly supported when the same survey method, denominator and cut-offs are used.</td>
    </tr>
    <tr>
      <td>Rising PGSI-based prevalence in Australia shows harm is increasing</td>
      <td>Consistent with several surveys, but the size of the increase depends on survey design. Trends within a single series are more reliable than comparisons between series.</td>
    </tr>
    <tr>
      <td>Screening tools are accurate at identifying clinical gambling disorder</td>
      <td>Partly supported. They identify people without problems well, but positive predictive power is modest in general populations.</td>
    </tr>
  </tbody>
</table>

Reading your own score

If you have completed a PGSI-based self-test, these points may help you read the result.

The score describes the past 12 months. It does not show how things are changing. A score that is lower than a year ago and a score that is higher than a year ago mean different things, even if they are the same number.

Look at the items, not only the total. Answers of "most of the time" or "almost always" to borrowing money, financial problems or chasing losses are meaningful on their own, whatever the total.

The category labels are approximate. The boundaries between categories are statistical conventions, not natural thresholds. A score of 7 and a score of 8 differ by one answer.

Context matters. Other people's experiences, mental health, finances and relationships are all relevant to how gambling is affecting a life. A trained counsellor or doctor can consider these in a way no questionnaire can.

The PGSI is valuable because it gives researchers and services a common language for gambling problems. It is most useful when read for what it is: a structured snapshot of the past year, sensitive to severity but not a substitute for a conversation with someone qualified to help.

Free, confidential support for anyone in Australia affected by gambling, including family members, is available 24 hours a day through the National Gambling Helpline on 1800 858 858 and through Gambling Help Online. If gambling is linked to thoughts of suicide or a crisis, contact Lifeline on 13 11 14 or emergency services on 000.