NUR3202C Research and Evidence-Based Healthcare Consolidation Lecture PDF

NUR3202C Research and Evidence-Based Healthcare Consolidation Lecture Asst. Prof Jocelyn Chew PhD, BSN (Hons), RN Assistant Professor, NUS N...

NUR3202C Research and Evidence-Based Healthcare Consolidation Lecture Asst. Prof Jocelyn Chew PhD, BSN (Hons), RN Assistant Professor, NUS Nursing Assistant Professor, Department for Biomedical Informatics, NUS Clinical Lead (Data Insights Unit), Deputy Chief Executive (Education & Research), NUHS Assistant Professor, Behaviour and Implementation Science Interventions (BISI), NUS ©©Copyright CopyrightNational NationalUniversity UniversityofofSingapore. Singapore.All AllRights RightsReserved. Reserved. 1  Quiz © Copyright National University of Singapore. All Rights Reserved. 2  Aims, objectives, questions, and hypotheses MUST BE CONSISTENT © Copyright National University of Singapore. All Rights Reserved. 3  Differences Aim: Broad, general, long-term E.g. To develop a psychoeducational program to improve diabetes self-care. Objectives: specific, focused, short-term, measurable 1 aim can have multiple objectives E.g. To test the effectiveness of a psychoeducational program to improve diabetes self-care among people with type 2 diabetes. Research questions: rephrase objectives to focus on variables E.g. What is the effectiveness of a psychoeducational program to improve diabetes self-care among people with type 2 diabetes? Making it sound like a question Research hypotheses (if applicable) For quantitative studies E.g. Patients receiving the psychoeducational program will exhibit a higher level of diabetes self-care as compared to patients who do not. Require null hypothesis and alternative hypothesis © Copyright National University of Singapore. All Rights Reserved. 4  Study designs based on research aims Qualitative Quantitative Testing the effectiveness of the study Descriptive Experimental (hypothesis testing) Phenomenological Randomized controlled trials Quasi-experimental trials Ethnographical Non-experimental (descriptive, Grounded theory correlational) Summarise/describe a certain observation Participatory action research Cross-sectional (PAR) Cohort Difficult to enroll patients in? Topic is taboo (unlikely for self-volunteer) Case-control Sensitive topics to find out more in-depth about it Can draw out subjective data and experience © Copyright National University of Singapore. All Rights Reserved. 5  Qualitative vs Quantitative Info comes frm bottom up (don’t prescribe info, wait for info/themes to come) Approach Qualitative Quantitative Focus Quality (features) Quantity (numbers) Reasoning Usually inductive Usually deductive Goal Understand Predict, test hypotheses Ensure that number Sample size Small, purposive allows findings to be generalized Large, general + instruments Data collection Interviews, observations Questionnaires, experiments Subjective findings Data analysis Researchers’ interpretation Statistical methods Difficult to manipulate Results/findings Usually verbatim quotes Usually precise numbers From interviewees © Copyright National University of Singapore. All Rights Reserved. 6  Quantitative research question PICO Examples Population &/problem Women undergoing menopause (characteristics of target group of people) Define specific group of subjectives— what characteristics u want ur population to have Intervention/exposure High intensity interval training (HIIT) (what do you want to do to the people?) What do you want to test? Comparison Aerobic exercise (not applicable for qualitative studies) Another existing intervention/standard treatment/nothing Outcome Menopausal symptoms including hot flashes, insomnia, (What is being measured to indicate effect of intervention/exposure?) and brain fog Variables that you are intending to measure Or ‘test’ > QUANTI study Research aim: To examine the effectiveness of HIIT on menopausal symptoms among women undergoing menopause. Research question: What is effectiveness of HIIT on menopausal symptoms including hot flashes, insomnia, and brain fog among women undergoing menopause? © Copyright National University of Singapore. All Rights Reserved. ‘Explore perception/experience’ > QUALI study 7  Qualitative research question NO Outcome PICo Examples Population &/ problem Patients with type 2 diabetes (characteristics of target group of people) Interest (phenomenon of interest Experience with diet restriction e.g. experience, behaviour, event) Context To improve blood glucose control (What is the setting?) Research question: What is the experience of patients with type 2 diabetes on diet restriction to improve blood glucose control? © Copyright National University of Singapore. All Rights Reserved. 8  Literature review © Copyright National University of Singapore. All Rights Reserved. 9  Literature review Systematic review Purpose Provide context/background Identifies, selects, synthesises, and appraises information, not meant to answer studies that meet prespecified inclusion research question. criteria to answer a research question. Protocol No protocol A-priori protocol is developed and published (PROSPERO) Search Nil, normally includes well-known Well-defined, comprehensive search strategy articles Methodological Nil Internal validity is judged by various tools eg appraisal ROB Synthesis Usually narrative Narrative, meta-analysis, meta-synthesis Findings Not reproducible Reproducible © Copyright National University of Singapore. All Rights Reserved. 10  Level of evidence Systematic reviews of systematic reviews Umbrella review Synthesized evidence Meta- QUANTI way of aggregating numbers in systematic review analyses Meta-synthesis: QUALI way of aggregating narrative data Systematic reviews STRONGEST GOLD STANDARD FOR ESTABLISHING EVIDENCE > well controlled research design that ensures final outcome is truly Randomized due to manipulation of intervention and nothing else (no co- Experimental studies (causal r/s) controlled trials variants/confounding factors contributing to outcome) Quasi-experimental Level of evidence studies Cohort studies Observational studies Cross-sectional studies Case-control studies Case reports Background information/expert opinion/textbooks/editorials © Copyright National University of Singapore. All Rights Reserved. 11  Just understand the logic of doing a systematic review: Steps to perform a systematic review Find a good topic Formulate clear and well-defined research question Develop systematic review protocol Conduct systematic search strategy TiAb and full-text screening using eligibility criteria Methodological appraisal Data extraction & organisation Data analysis Evidence quality appraisal Write: integrate, synthesize, summarize © Copyright National University of Singapore. All Rights Reserved. 12  Qualitative research study design © Copyright National University of Singapore. All Rights Reserved. 13  Introduction Definition: A type of research method that collects non- numerical data for in-depth understanding of phenomenon in their natural setting. Purpose: Explore a phenomenon (e.g. perception, meaning, experience) that is vague Groundwork for quantitative study when there is insufficient insights E.g. why people behaviour a certain way? Explain a quantitative result © Copyright National University of Singapore. All Rights Reserved. 14  Formulating research question Types of inquiry Examples Ontological (understand “What is the lived experiences of cancer participants’ realities) survivors?” Epistemological (understand “What are the motivators of heart failure knowledge of phenomenon) self-care?” Refer to previous week 2 lecture: PIO © Copyright National University of Singapore. All Rights Reserved. 15  Qualitative designs Qualitative designs Description Descriptive (most Describe and interpret perceptions/meanings. basic) Grounded theory Collect rich data on a topic to inductively develop theories. Phenomenological Understand a phenomenon by describing and How does person interpreting participants’ lived experiences. perceive world? Ethnography Researchers immerse themselves in target groups to understand culture. Participatory action Both researchers and participants conduct research together to drive social change. © Copyright National University of Singapore. All Rights Reserved. 16  Data collection methods In-depth interviews Individual vs focus group Providing context of how the interviewee is like that might help the research Semi-structured vs Unstructured Observations + field notes Use of 5 senses Surveys with open-ended questions Secondary data Existing texts, images, audio-recordings, video-recordings © Copyright National University of Singapore. All Rights Reserved. 17  In-depth interview techniques Aim: Evoke thick and rich responses to obtain in-depth information Build rapport + participant information Anonymity & confidentiality Permission to audio-tape record Develop an interview guide with open-ended questions More Why? How? Talk less, listen more  use prompts, silence DO NOT use leading questions © Copyright National University of Singapore. All Rights Reserved. 18  Sampling methods QUALI STUDY Sampling methods Description Convenience Volunteers through advertisements Purposive Non-probability sampling based on criteria set beforehand Not relying on randomisation/chance, dependent on researcher’s subjective view on who is suitable Usually for more taboo/ Snowball Recruited participants to recommend others sensitive topics Theoretical Decide on next target participant as collection (grounded-theory) continues Changing criteria based on current data analysis © Copyright National University of Singapore. All Rights Reserved. 19  Sample size Sample size normally based on data saturation: when no more new information emerges Qualitative designs Sample size guide Descriptive (most basic) >12 (Clark & Braun, 2013) Grounded theory 20-30 Phenomenological ~10 Ethnography 25-50 Focus group ≥3 groups, each 7-10 participants © Copyright National University of Singapore. All Rights Reserved. 20  Qualitative data analysis and trustworthiness © Copyright National University of Singapore. All Rights Reserved. 21  Basic data analysis methods Approach Content analysis Thematic analysis Purpose To identify & describe To identify & interpret common ideas meaning/essence Nature of data Manifest Latent Nature of findings Descriptive Interpretive Using own subjective understanding of data to derive themes (more abstract themes) Data analysis process Usually deductive Usually inductive Interest What Why © Copyright National University of Singapore. All Rights Reserved. 22  General data analysis steps 1. Prepare: Materials for data analysis Transcript (include context of data collection) Data must be clean before being analysed (QUALI and QUANTI) e.g. situation (time and date), environment (private room or in the open space), facial expression (e.g. facial grimace when talking about sensitive issues) 2. Immerse/familiarise: Iterative reading 3. Code: Label patterns/meaning units 4. Allow themes & subthemes to emerge © Copyright National University of Singapore. All Rights Reserved. 23  6-steps thematic analysis (Braun & Clarke, 2006) 1. Familiarize with data Transcribing data, reading and rereading the data, noting down initial ideas 2. Generating initial Coding interesting features of the data in a systematic fashion across the entire data set, collating codes data relevant to each code 3. Searching for themes Collating codes into potential themes, gathering all data relevant to each potential theme Checking if the themes work in relation to the coded extracts (Level 1) and the entire data set 4. Reviewing themes (Level 2), generating a thematic ‘map’ of the analysis 5. Defining and naming Ongoing analysis to refine the specifics of each theme, and the overall story the analysis tells, themes generating clear definitions and names for each theme Selection of vivid, compelling extract examples, final analysis of selected extracts, relating back of 6. Producing the report the analysis to the research question and literature, producing a scholarly report of the analysis © Copyright National University of Singapore. All Rights Reserved. 24  Manual coding Maintaining social life Maintaining meaning in Maintaining passion life – Hopefulness Grateful for one more day Integrating CHF restrictions into new life © Copyright National University of Singapore. All Rights Reserved. 25  Computer-assisted data analysis E.g. Nvivo, Atlas.ti, MaxQDA © Copyright National University of Singapore. All Rights Reserved. 26  Trustworthiness A set of strategies used to establish trust or confidence (Lincoln, 1989; Morse, 2015) How sure that we are exploring what How trustworthy/true/ Generalisability of results, we are supposed to explore? objective are your findings? how well can we use the same findings on another population with similar Criterion Quantitative Qualitative ‘Rigor’ characteristics? Truth value Internal validity Credibility Applicability External validity Transferability Consistency Reliability Dependability Neutrality Objectivity Confirmability How objective it is © Copyright National University of Singapore. All Rights Reserved.  Rigor in qualitative studies Criterion Explanation Some strategies Credibility Confidence in the truth of Prolonged the findings engagement Ensure an accurate Member checking description or Interview technique interpretations of human Reflexivity journal experience that individuals More than one data with similar experiences analyst would recognize these descriptions Transferability Extent to which a reader Thick description of (Krefting, 1991; Morse, 2015) can transfer the findings to the sample, setting, another similar situation or and context context (Krefting, 1991; Morse, 2015) © Copyright National University of Singapore. All Rights Reserved.  Rigor in qualitative studies Criterion Explanation Some strategies Dependability Stability of the findings Dense description of across time research methods Ensure that research can be Triangulation audited; variations can be Peer examination traced back to identifiable Audit trail of sources decision-making process Confirmability Stability of the findings Triangulation across contexts and Reflexivity population Peer debriefing (Krefting, 1991; Morse, 2015) Neutrality of the data as opposed to the researcher (Krefting, 1991; Morse, 2015) © Copyright National University of Singapore. All Rights Reserved.  Quantitative research study design © Copyright National University of Singapore. All Rights Reserved. 30  Key dimensions to consider: Experimental vs non-experimental (RCT, quasi-experimental to identify causal r/s) vs (e.g. descriptive, correlational, comparative) Cross-sectional vs longitudinal Snapshot vs change over time Retrospective vs prospective Using past data Collecting and using future data (new) © Copyright National University of Singapore. All Rights Reserved. 31 SUMMARY © Copyright National University of Singapore. All Rights Reserved. 32  True experimental research (RCT) Gold standard for testing causal relationships Also called pretest-posttest design with randomization Non-equivalent pretest-posttest is called quasi-experimental trial If there is NO RANDOMISATION = Quasi-experiment Characteristics: Intervention: manipulation of IV Control group Random assignment (group assignment by equal chance to eliminate confounding factors, allowing us to ascertain that DV is indeed caused by IV) DIFFERENT from random sampling Randomising > controlling all other confounding factors © Copyright National University of Singapore. All Rights Reserved. 33  Randomization Minimise selection bias through allocation concealment Trials with unclear randomization shown to overestimate interventional effects by 40% Of type 1 error (Schulz & Grimes, 2002) Best to have different people performing different steps of randomization to prevent bias during Prevent influence of results/outcome Participant recruitment Participant allocation Intervention administration Outcome assessment © Copyright National University of Singapore. All Rights Reserved. 34  Would be the only variable/intervention that caused him to lose weight RCT: cause-and-effect Pre-test Intervention group 6 months Weight measurement Intervention Weight measurement Random assignment Control group NO INTERVENTION Weight measurement Weight measurement Pre-test © Copyright National University of Singapore. All Rights Reserved. Understand concepts behind each experiment © Copyright National University of Singapore. All Rights Reserved.  One group pretest-posttest design Compare between results Control is pre-test SAME GROUP OF PEOPLE © Copyright National University of Singapore. All Rights Reserved.  Non-experimental research Descriptive Reporting descriptive statistics E.g. Examine the quality of life among patients with CHD Correlational Analysis/association/rs between factors Examine relationship between variables E.g. Examine the relationship between medication adherence and quality of life in patients with CHD Comparative Compare differences between 2 groups, NO intervention (hence non-experimental) To compare variables between samples E.g. A comparative study on health-related quality of life between patients with MI and DM © Copyright National University of Singapore. All Rights Reserved.  Cross-sectional vs longitudinal © Copyright National University of Singapore. All Rights Reserved. 39  Retrospective vs prospective © Copyright National University of Singapore. All Rights Reserved. 40  Quantitative sampling © Copyright National University of Singapore. All Rights Reserved. 41  Common statistical terms Population: Collection of entire set of individual objects or events of interest. e.g. census data (information recorded about the population of well-defined geographical areas), hospital database Parameter: Numerical characteristic of population. Variable: Characteristic that is being measured. Sample: Representative subset of a population. To make estimates or test hypotheses about population data when infeasible to collect population data. Statistic: measure that describes the sample © Copyright National University of Singapore. All Rights Reserved. 42  Why do we use sampling in clinical research? Ideal case: look at every single person Impossible due to logistics, cost and ethical limitations Next best case: look at a representative subset of the population Factors that influence representativeness of sample How clearly the population and sample is defined Sampling method Sample size © Copyright National University of Singapore. All Rights Reserved. 43  Sampling Ideally, sample should be derived through probability sampling (randomly selected to ensure representativeness) reduces risk of sampling bias and enhance internal and external validity. non-probability sampling methods for convenience and affordability. Sampling bias: due to systematic errors in sampling process (preventable) Sampling bias: e.g. difference between RGPS mean PSLE score and national mean PSLE score. Sampling error: difference between population parameter and sample statistic due to chance (inevitable) e.g. difference between mean PSLE score in people staying in Clementi and national mean PSLE score. © Copyright National University of Singapore. All Rights Reserved. 44  Why do you need sample size calculation? To ensure that the sample is a sufficient representation of the population, to have sufficient statistical power to detect a true effect when there is one. Too many participants  overpowered  redundant, waste resources Too few participants  underpowered  risk of excessive selection (type II error); inaccurate; unreliable Mandatory for ethics application, grant application and research publications. To ensure exposure to any potential risk is limited to as few subjects as possible To ensure reasonable chance of identifying effect if there is one © Copyright National University of Singapore. All Rights Reserved. 45  Sampling methods Probability Non-probability sampling sampling Simple Convenience random Stratified Consecutive random systematic Snowball © Copyright National University of Singapore. All Rights Reserved. 46  Data collection © Copyright National University of Singapore. All Rights Reserved. 47  Key considerations Sample size Consistency/fidelity in recruitment and data collection method (training process, qualification, inter-rater reliability) Control over extraneous factors (e.g. noise, temperature, lighting) © Copyright National University of Singapore. All Rights Reserved. 48  Data collection instruments Self-report surveys: open-ended, questionnaires, scales (Likert scale, visual analog scale) Prone to response set biases: social desirability bias, extreme response set bias, acquiescence response set bias. Observations (unstructured/structured) Biophysiological measures © Copyright National University of Singapore. All Rights Reserved. 49 © Copyright National University of Singapore. All Rights Reserved. 50  Levels of measurement © Copyright National University of Singapore. All Rights Reserved. 51  Levels of variable measurement Level of measurement: Classification of variable measurements Levels of variable Numerical Categorical (quantitative) (qualitative) Nominal Ratio (continuous) Interval (Discrete) Ordinal (Ordered) (Without order) © Copyright National University of Singapore. All Rights Reserved. 52 © Copyright National University of Singapore. All Rights Reserved. 53  A variable has one of four different levels of measurement Nominal – Latin. “name only” Yes/no, TDH (no meaningful order) Ordinal – Ordered/ranked levels Rank 1st, 2nd, 3rd (have meaningful order) Interval – Intervals fixed btw all Rating options but no true zero Ratio – Interval but with Scale: 0-∞ true/absolute/meaningful zero © Copyright National University of Singapore. All Rights Reserved.  Reliability and validity © Copyright National University of Singapore. All Rights Reserved. 55 © Copyright National University of Singapore. All Rights Reserved. 56  Before using a tool, ensure validity & reliability Validity (content & criterion validity) Extent to which an instrument measures what it is supposed to measure in a specific population Reliability (α, ICC/k, ICC) Extent to which results obtained are stable across items, raters & time Which one should be established first? Can one exist without the other? © Copyright National University of Singapore. All Rights Reserved.  No instrument can measure a construct perfectly: Measurement error Obtained results = True result ± Error © Copyright National University of Singapore. All Rights Reserved. 58  Sources of measurement error Situational & Transient Response environmental personal bias factors factors Researcher: friendly/RBF Social desirability Mood, hunger, business Location: private/open Acquiescence Environment: bright, time of Recall day, temperature, humidity Attrition Train researchers to collect data Prep participant, ensure Assure confidentiality, according to protocol and try to readiness to respond anonymity, time ensure all else same © Copyright National University of Singapore. All Rights Reserved.  Reliability Stability Extent to which score stays the same over time Test-retest reliability (ICC) Internal consistency Extent to which all other items within the survey measures the same trait. Cronbach’s alpha (α) Equivalence Extent to which 2 or more independent observers agree on scores Categorical measures: Cohen’s Kappa(κ) (≥0.6) Continuous measures: ICC (≥0.7) © Copyright National University of Singapore. All Rights Reserved. 60  Validity-Content validity Appropriately and adequacy of instrument content Judged by expert panel © Copyright National University of Singapore. All Rights Reserved. 61 © Copyright National University of Singapore. All Rights Reserved. 62  Validity-Criterion-related validity Extent to which instrument corresponds to gold standard or another well-established measure of target variable Types: Concurrent validity: correlation of tested and gold standard instrument scores at the same time (r) Predictive validity: correlation of scores with of outcome (r) © Copyright National University of Singapore. All Rights Reserved. 63  Descriptive statistics © Copyright National University of Singapore. All Rights Reserved. 64  Definition Inferential statistics: make generalisations & predictions on a population based on sample data/results Descriptive statistics: Make descriptions & summaries of central tendency (mean, median, mode), data spread (variance, standard deviation, range, IQR), count (N/n), proportion (%), skewness, kurtosis etc. Recap: Population mean = μ Sample mean = x̄ © Copyright National University of Singapore. All Rights Reserved. 65  Types of statistics Types of statistics Descriptive Inferential statistics statistics Continuous variables Categorical variables Measure of Measure of frequency Measure of central dispersion/variability Hypothesis testing Regression analysis tendency (count, %, cross- (standard deviation, tabulations, graphs, (e.g. z-test, t-test, f- (e.g. linear regression, (mean, median, mode) variance, range, illustrations) test, x2 test, ANOVA) logistic regression) interquartile range) © Copyright National University of Singapore. All Rights Reserved. 66  Descriptive statistics (Categorical variables; measure of frequency) Frequency (%): e.g. Males 70 (70%); females 30 (30%) Nursing year 1: 20 (20%); year 2: 30 (30%); year 3: 25 (25%); year 4: 25 (25%) Cross-tabulation Year 1 Year 2 Year 3 Year 4 Total Males 15 (21.4%) 13 (18.6%) 19 (27.1%) 5 (7.1%) 70 Females 5 (16.7%) 17 (56.7%) 6 (20.0%) 20 (66.7%) 30 Total 20 (20%) 30 (30%) 25 (25%) 25 (25%) 100 © Copyright National University of Singapore. All Rights Reserved. 67  Descriptive statistics (Categorical variables; measure of frequency) https://medium.com/@Lynia_Li/as-you-know-there-are-many-types-of-charts-to-be-used-in-data-visualization-54da9b97092e https://datavizcatalogue.com/methods/pictogram.html#google_vignette © Copyright National University of Singapore. All Rights Reserved. 68  Descriptive statistics (Continuous variables; measure of central tendency) Statistic Formula Example Mean Example: Consider the data set {2, 4, 6, Add all values divided by total number of (arithmetic average of a set of values) 8, 10}. values: (more suitable for symmetric (2+4+6+8+10/ 5) distribution, often reported with Mean = 5 x̄=∑ x/n standard deviation e.g. mean (SD) ) Median (middle value of a data set when 𝑛𝑛+1 th arranged in ascending or descending If n is odd: term Example: Data set {3, 6, 9, 12, 15}. 2 𝑛𝑛 th 𝑛𝑛+1 th order) (more suitable for skewed term+ term Median = 9 If n is even: 2 2 distribution; often reported with 2 interquartile range e.g. median (IQR)) Mode Value that appears most frequently in a data Example: Data set {2, 3, 4, 4, 5, 6, 6, 6, 7}. (value that occurs most frequently in a set Mode = 6 data set) © Copyright National University of Singapore. All Rights Reserved. 69  Descriptive statistics (Continuous variables; measure of central tendency) Histogram Range Range Interquartile range (IQR) Interquartile range (IQR) Q1 Q2 Q3 Q1 Q2 Q3 © Copyright National University of Singapore. All Rights Reserved. 70  Descriptive statistics (Continuous variables; measure of central tendency) Boxplot © Copyright National University of Singapore. All Rights Reserved. 71  Normality of distribution All kinds of naturally occurring variables are usually normally distributed (Bell curve). E.g. height, weight, job satisfaction Because of how common a normal distribution is, many statistical tests are designed for normally distributed populations. Main property: mean, median and mode are the same (symmetrical curve) Skewness: lack of symmetry tells us direction of variability different from variance, which tells magnitude of variability Negative skew Positive skew © Copyright National University of Singapore. All Rights Reserved. 72  Descriptive statistics (Continuous variables; measure of dispersion/variability) Variance: average of squared differences of each datapoint from mean (squared unit of mean) σ2=0: all data values are the same Small variance=data are close to mean and each other Large variance=data are far from mean and each other Standard deviation: square root of variance (same unit as mean) © Copyright National University of Singapore. All Rights Reserved. 73  Empirical rule aka 68-95-99.7 rule tells you where most of your values lie in a normal distribution: ~68% of values are within 1 SD from the mean ~95% of values are within 2 SD from the mean ~99.7% of values are within 3 SD from the mean © Copyright National University of Singapore. All Rights Reserved. 74  Inferential Statistics © Copyright National University of Singapore. All Rights Reserved. 75  Purpose To generalise sample characteristics to population parameters that are often unknowable Generalisation are still just estimations and we have to account for inaccuracies and errors using confidence interval (CI) CI: A range of values where the true mean lies Normally present mean, SD and CI © Copyright National University of Singapore. All Rights Reserved. 76  Types of hypotheses Null hypothesis: no significant relationship between independent and dependent variables, no difference between groups Research hypothesis: Hypothesis of point-prevalence: The behaviour change program significantly improve weight loss among patients with obesity by 5%. Hypothesis of difference: There is significant difference in weight loss among patients who receive the behaviour change program (intervention group) as compared to those who do not (control group) Hypothesis of association: There is significant relationship between self-efficacy and dietary control among patients with obesity. © Copyright National University of Singapore. All Rights Reserved. 77  Hypothesis testing Steps: 1. State null hypothesis (H0) H0: μ = m0 (which means no effect, no relationship, no difference between true and observed mean) 2. State alternative hypothesis(es) (H1) H1: μ ≠ m0 (two-tailed: there is difference between true and observed mean) H2: μ > m0 (upper-tailed: true mean > observed mean) H3: μ < m0 (lower-tailed: true mean < observed mean) 3. Set significance level (α) Normally 0.05, which means that there is 5% chance that you will reject your H0when H0 is true 4. Collect data 5. Calculate statistics including p-value (probability of observing a sample statistic by chance, given that H0 is true) E.g. if p-value=0.03, it means that 3 out of 100 times of your sample observation occurs by chance, given that H0 is true (meaning that it is unlikely that your observation occurred by chance) If p-value=0.90, it means that 90 out of 100 times of your sample observation occurs by chance, given that H0 is true (meaning that it is very likely that your observation occurred by chance) 6. Interpret results © Copyright National University of Singapore. All Rights Reserved. 78  Types of errors in hypothesis testing If null hypothesis (H0) True False is: Type I error True positive Rejected False positive Type II error Not rejected True negative False negative © Copyright National University of Singapore. All Rights Reserved. 79  Parametric & non-parametric tests Parametric Non-parametric Distribution Follows a normal No need to follow normal distribution assumed (Gaussian) distribution (mostly based on rank order or how common data is) Central tendency Assess group mean Assess group median measure Types of variables Continuous variables Can be used for both continuous and discrete variables Statistical power Higher Lower © Copyright National University of Singapore. All Rights Reserved. 80  Common assumptions for parametric tests DV is continuous (interval/ratio) (e.g. age, IQ, height) DV follows a normal distribution Homogeneity of variance between groups (FYI: Levene's Test of Equality of Variances) Comparison groups are independent (subjects in both groups cannot influences each other) Preferably no significant outliers © Copyright National University of Singapore. All Rights Reserved. 81  Checking for normality 1. Visualization: https://towardsdatascience.com/6-ways-to-test-for-a- normal-distribution-which-one-to-use-9dcf47d8fa93 https://blogs.sas.com/content/iml/2019/07/22/extreme- Q-Q plot Histogram (if presenting to laymen) value-normal-data.html 2. Statistical hypothesis testing: Shapiro-Wilk test © Copyright National University of Singapore. All Rights Reserved. 82  Comparing: Parametric Non-parametric Examples (parametric) (e.g. BMI) (e.g. pain score) One group One sample t- Wilcoxon Signed Is the mean age of menopause in Singaporean mean/median test/z-test Rank test women 49 years old? Two means/medians Paired sample t- Wilcoxon Signed Would there be a significant change in blood from the same test Rank test glucose level of the intervention group between person baseline and one-year follow-up? Two independent Independent Mann-Whitney U test Are there significant differences in anxiety group means/medians samples t-test levels between undergraduate nursing students in year 1 compared to those in year 4? ≥Three independent One-way Analysis Kruskal-Wallis test Are there significant differences in BMI among group means/medians of variance adults who underwent bariatric surgery, (ANOVA) exercise intervention, diet intervention or no intervention (control group)? ≥Three independent Two-way ANOVA Friedman’s test Are there significant differences in blood group means/medians (e.g. repeated pressure between those with high, moderate of two independent measure ANOVA) and low level of physical activity at 3-month, 6- variables (categorical) month and 9- month follow-up? © Copyright National University of Singapore. All Rights Reserved. 83 © Copyright National University of Singapore. All Rights Reserved. 84  z-test vs t-test Used to compare means between two groups/conditions Measures exactly how many standard deviations above or below the mean a data point is. © Copyright National University of Singapore. All Rights Reserved. 85  z-test Only used when population SD is known to estimate population mean (in real-life, its unlikely to know population SD and not mean and sample SD is may not be a good estimate). Hence if not known, use t-test. Uses standard normal distribution where population mean (μ)=0; population SD (σ)=1. © Copyright National University of Singapore. All Rights Reserved. 86  t-test T-distribution also defines population mean (μ)=0 but SD varies with sample size © Copyright National University of Singapore. All Rights Reserved. 87  Standard Normal Distribution vs student T distribution The larger the sample size, the more similar (>30 somewhat identical) © Copyright National University of Singapore. All Rights Reserved. 88  F-test (ANOVA) Used to compare variances or equality of means among three or more groups/conditions © Copyright National University of Singapore. All Rights Reserved. 89  ANOVA Analyses how entire set of group means are spread out regardless of group differences (assuming no true differences between groups) Results in F ratio/statistic: estimate of the variability between groups as compared to that within groups © Copyright National University of Singapore. All Rights Reserved. 90  Measures of linear association e.g. bivariate correlation, chi-square test and regression © Copyright National University of Singapore. All Rights Reserved. 91  Introduction Produces a coefficient used to quantify the strength and direction of a relationship/association between two or more variables Value of coefficient ranges from -1 to +1. Note: coefficient is entirely separate from statistical significance. i.e. it is plausible for a strong association to be non-statistically significant and vice versa. © Copyright National University of Singapore. All Rights Reserved. 92  Bivariate correlation Note: not applicable to curvilinear or discontinuous relationships (violates assumptions) © Copyright National University of Singapore. All Rights Reserved. 93  Chi-square test of independence Measures significance of an association between two categorical variables E.g. Among patients with Hodgkin's disease, is there a significantly disproportionate number of females than males, indicating a significant association between haematological cancer and sex? © Copyright National University of Singapore. All Rights Reserved. 94  Linear regression An estimation of the association between a continuous DV and ≥1 IV Assumptions: Simple linear Linear relationship regression Independence Homoscedasticity Multiple linear Regression regression Normality Nonlinear regression © Copyright National University of Singapore. All Rights Reserved. 95  Simple linear regression 𝑦𝑦 = a + b𝑥𝑥 + 𝑒𝑒 E.g. Lung function Forced expiratory volume= 0.0125 + 0.364(age) + 0.0135 © Copyright National University of Singapore. All Rights Reserved. 96  Multiple linear regression 𝑦𝑦 = a + b1𝑥𝑥1 + b2𝑥𝑥2+b3𝑥𝑥3+…+ e E.g. Lung function (Ma et al., 2013) Forced expiration volume= -2.63 + 0.108(age) + 0.011(weight) + 0.021(height) + 0.738 © Copyright National University of Singapore. All Rights Reserved. 97  Logistic regression An estimation of the association between a binary DV and ≥1 IV © Copyright National University of Singapore. All Rights Reserved. 98  Comparisons Correlation (r) Regression (𝜷𝜷) Purpose Describe, infer Predict Parametric Pearson’s correlation coefficient - (aka Pearson’s 𝜌𝜌) Non-parametric Spearman rank-order correlation - coefficient (aka Spearman’s 𝜌𝜌) Effect size strength (Cohen’s Very weak: r < 0.3 standard) Weak: 0.3 < r < 0.5 Moderate: 0.5 < r < 0.7 Strong: r > 0.7 Examples (parametric) Is there a statistically significant Are genetics, stress, social support relationship between stress level and depression significant and blood pressure? predictors of weight status? © Copyright National University of Singapore. All Rights Reserved. 99