Class Notes (810,488)
United States (314,224)
Health (50)
HLTH 200 (18)
all (1)

[COMPLETE] HLTH 200 Lecture Notes 4.0 GPA Students

64 Pages
Unlock Document

University of Maryland
HLTH 200

Part I: Foundations (9/6-9/18) 12/9/13 6:33 PM Sessions 2-7: Foundations of Research • Learning Objectives o Understand the basic language of research o Understand the underlying philosophical issues that drive the research endeavor o Describe the types of research studies and types of relationships o Explain critical ethical issues that affect the researcher, research participant, and the research efforts o Understand where research problems come from o Learn how to develop a research question and hypothesis • Reading: Trochim Ch. 1 – Foundations o The language of research ! Research vocabulary ! Types of studies ! Time in research ! Variables ! Types of relationships " The nature of a relationship " Patterns of relationships ! Hypotheses ! Types of data ! The unit of analysis o The rationale of research ! Structure of research " Components of a study ! Deduction and induction ! Two research fallacies o Validity of research o Ethics in research ! The language of ethics o Conceptualizing ! Where research topics come from ! Feasibility ! The literature review Part II: Sampling (9/20-9/25) 12/9/13 9:23 PM Sessions 8-10: Sampling • Learning Objectives o Understand external validity o Understand sampling terminology o Understand statistical terms in sampling o Understand probability sampling o Understand nonprobability sampling o Understand how and why different types of research questions require different types of samples • Reading: Trochim Ch. 2 – Sampling o External validity ! Threats to external validity ! Improving external validity o Sampling terminology o Statistical terms in sampling ! The sampling distribution ! Sampling error ! The 65, 95, 99 percent rule o Probability sampling ! Some definitions ! Simple random sampling ! Stratified random sampling ! Systematic random sampling ! Cluster (area) random sampling ! Multistage sampling o Nonprobability sampling ! Accidental, haphazard, or convenience sampling ! Purposive sampling " Modal instance sampling " Expert sampling " Quota sampling " Heterogeneity sampling " Snowball sampling Part III: Measurement (9/27-10/18) 12/10/13 12:17 AM Sessions 11-12: Theory of Measurement • Learning Objectives o Be able to assess the quality of a measurement tool o Understand the theory of what constitutes good measure o Understand the consistency or dependability of measurements o Understand the four major levels of measurement • Reading: Trochim Ch. 3 – The Theory of Measurement o Construct validity ! Measurement validity types " Translation validity " Criterion-related validity ! Idea of construct validity ! Convergent and discriminant validity " Convergent validity " Discriminant validity " Putting it all together ! Threats to construct validity " Inadequate preoperational explication of constructs " Mono-operation bias " Mono-method bias " Interaction of different treatments " Interaction of testing and treatment " Restricted generalizability across constructs " Confounding constructs and levels of constructs " The social threats to construct validity ! Reliability " True score theory " Measurement error • What is random error? • What is systematic error? • Reducing measurement error " Theory of reliability " Types of reliability • Inter-rater or inter-observer reliability • Test-retest reliability • Parallel forms reliability • Internal consistency reliability " Reliability and validity ! Levels of measurement " Why is level of measurement important? Sessions 13-17: Survey Research • Learning Objectives o Understand how to construct a research survey o Understand how to construct questions used for survey research o Explain the role of an interviewer o Describe the major types of surveys o List advantages and disadvantages of survey methods o Know the steps involved in conducting a community survey • Reading: Trochim Ch. 4 – Survey Research o Constructing the survey ! Types of questions ! Question content ! Response format ! Question wording ! Question placement ! The golden rule o Interviews ! The role of the interviewer ! Training the interviewers ! The interviewer’s kit ! Conducting the interview o Surveys ! Types of surveys ! Selecting the survey method ! Advantages and disadvantages of survey methods Session 14: Scales and Indices • Learning Objectives o Distinguish between a scale and an index o Explain the difference between multidimensional and uni- dimensional scaling o Understand the three types of uni-dimensional scales o Learn how to use indexes and scales o Learn when each type of index or scale is most appropriate • Reading: Trochim Ch. 5 – Scales and Indexes o Indexes ! Some common indexes ! Constructing an index o Scaling ! General issues in scaling " Purposes of scaling " Dimensionality " Uni-dimensional v. multi-dimensional ! Thurstone scaling ! Likert scaling ! Guttman scaling o Indexes and scales Sessions 18-20: Qualitative and Unobtrusive Measures • Learning Objectives o Understand the purpose of a qualitative study o Describe qualitative data and how it differs and complements quantitative data o Explain different types of qualitative methods o Explain standards for judging the validity of qualitative measurement • Reading: Trochim Ch. 6 – Qualitative and Unobtrusive Measures o Qualitative measures ! When to use qualitative research " Generating new theories or hypotheses " Achieving deeper understanding of the phenomenon " Developing detailed stories to describe a phenomenon " Mixed methods research ! Qualitative and quantitative data " All qualitative data can be coded quantitatively " All quantitative data is based on qualitative judgment ! Qualitative data ! Qualitative traditions " Ethnography " Phenomenology " Field research " Grounded theory ! Qualitative methods " Participant observation " Direct observation " Unstructured interviewing " Case studies ! The quality of qualitative research " Credibility " Transferability " Dependability " Confirmability o Unobtrusive measures ! Indirect measures ! Content analysis ! Secondary analysis of data Part 4: Design (10/23-11/11) 12/9/13 6:30 PM Sessions 22-23: General Design Issues • Learning Objectives o Understand internal validity closely connected to research design o Understand major threats to internal validity o Learn to classify the major types of designs o Understand the relationships among designs and their importance when making design choices • Reading: Trochim Ch. 7 - Design o Internal validity ! Establishing cause and effect " Temporal precedence " Covariation of the cause and effect " No plausible alternative explanations ! Single-group threats ! Multiple-group threats ! Social interaction threats o Introduction to design o Types of designs Sessions 25-26: Experimental Design • Learning Objectives o Understand the idea and purposes of experimental design o Describe why experimental design is strong in internal validity o Understand the key distinguishing feature of experimental design o Distinguish between random selection and random assignment o Describe how to classify the different experimental designs • Reading: Trochim Ch. 8 - Experimental Design o Introduction to experimental design ! Experimental designs and internal validity ! Two-group experimental designs ! Probabilistic equivalence ! Random selection and assignment o Classifying experimental designs o Factorial designs ! The basic 2x2 factorial design " The null outcome " The main effects " Interaction effects ! Factorial design variations " A 2x3 example " A three-factor example " Incomplete factorial design ! Randomized block designs ! Covariance designs ! Switching replications experimental designs Sessions 27-28: Quasi-Experimental Design • Learning Objectives o Understand the idea of quasi-experimental designs o Understand two of the classic quasi-experimental designs: the nonequivalent-groups design and the regression-discontinuity design o Understand the assortment of other quasi-experiments • Reading: Trochim Ch. 9 - Quasi-Experimental Design o The nonequivalent-groups design ! The basic design ! The bivariate distribution " Possible outcome 1 " Possible outcome 2 " Possible outcome 3 " Possible outcome 4 " Possible outcome 5 ! The regression-discontinuity design " The basic RD design • The role of the comparison group in RD designs • The internal validity of the RD design " Statistical power and the RD design " Ethics and the RD design ! Other quasi-experimental designs " The proxy-pretest design " The separate pre-post samples design " The double-pretest design " The switching replications design " The nonequivalent dependent variables (NEDV) design • The pattern-matching NEDV design " The regression point displacement (RPD) design Sessions 29-30: Designing Designs • Learning Objectives o Learn how to tailor a research design to fit the particular needs of the research context o Understand how to minimize the relevant threats to validity o Understand the logic of design construction o Learn how to design a research design o Clarify some of the basic principles of design logic • Reading: Trochim Ch. 10 - Designing Designs o Designing designs for research ! Minimizing threats to validity ! Building a design " Basic design elements " Expanding a design ! A simple strategy for design construction ! An example of a hybrid design Session 22: Introduction to Design and Analysis 12/9/13 6:30 PM There are 4 main steps to the research process. • Step 1: preliminary/formative o Formulate the research question o Develop hypotheses o Assemble team o Think through the design, rationale, alternatives, and constructs o Choose sample o Pilot tests (with IRB approval) • Step 2: final decision-making o Write up the design o Decide on the design, measures, and the sample o Finalize other details o Submit for IRB approval • Step 3: active data collection o Submit for IRB approval o Gather data • Step 4: analyses and write-up o Write up analytic plan o Prepare data o Analyze data o Write up data o Submit and publish data Introduction to design: • Design is used to structure the research activities and how all the major parts of the research project work together. The write-up of your design plan should show how all the major parts of the research project work together to address the question. It should also address rationale and alternatives that you abandoned, and maximize internal validity. • Elements of design o Observations and measures (notated as O): when stacked vertically on top of each other, O is collected at the same time o Treatments or programs (notated as X) o Groups (given their own lines) o Assignment to a group (describes how the group was assigned) ! Random assignment (R) ! Nonequivalent groups (N) ! Assignment by cutoff (C) o Time (from left to right) Session 23: Internal Validity, Cause and Effect, Threats 12/9/13 6:30 PM There are 3 classes of research design, and notation helps show common design substructures across different designs. • Post-test only randomized experiment: R X O; R O • Pre-post nonequivalent groups quasi-experiment: N O X O; N O O o Nonequivalent because you do not explicitly control the assignment, and the groups may not be similar to each other • Post-test only non-experiment: X O Internal validity is the approximate truth about inferences regarding cause-effect (causal) relationships. The primary consideration in internal validity is establishing cause and effect. The key question of internal validity is whether or not observed changes (effect) can be attributed to the program or intervention (cause) and not some other possible (alternative) cause. • Internal validity is only relevant to the specific study in question, so it isn’t concerned with generalizability (“zero-generalizability concern”) • Single-group threats to internal validity are criticisms that apply when you are only studying a single group. The single- group design goes as follows: administer program and measure outcome (X O); measure baseline, administer program, and measure outcome (O X O). The following threats will follow this example: The study of an education program for math for first graders; pre-post single group design. Measures (O) are standardized achievement tests, administered at the start of first nd grade and then at the start of 2 grade, with 2 forms (A and B). Adding a control group is one of the best ways to eliminate all single-group threats. o History threat: any event other than the program (cause) affects the outcome (post-test); specific extraneous factors (example: Participants pick up math concepts from watching Sesame Street, so their math skills improve anyway) o Maturation threat: natural development between pre- and post-test affects outcome (example: Participants would have learned these anyway even without the program, just because they got older). In the single-group design, when measuring things that could be affected by this, don’t wait too long between measurements. o Testing threat: the effect on the post-test of simply taking the pre-test; only occurs in pre-post designs (example: The pre-test might have had “priming effects” – that is, the kids might have learned something from the pre-test, not the post-test). Solve for this threat by using things from form A on pre-test and form B for post-test, and be careful of instrumental/parallel forms of reliability o Instrumentation threat: any change in the test from pre- test and post-test; only occurring in pre-post designs (example: Change is due to the different forms of the test (form A for pre-test, and form B for post-test) and not due to the program) o Mortality threat: (called attrition bias/study dropout in terms of Public Health) the nonrandom dropout between the pre-test and the post-test. Participants with low pretest scores drop out of the program (example: The kids with the lowest scores drop out, so they aren’t present in the post- test. Results won’t be able to determine if the math program improves something if it is only measuring the outcomes of the highest scoring students). Estimate the degree of mortality threat by comparing the dropout group against the non-dropout group on pretest scores. o Regression threat: when the sample is a nonrandom subgroup of a population, the average of the sample of the pre-test “regresses” toward the population mean on the post- test, always happens statistically. “Outliers” tend to not stay outliers for log – they regress towards the mean. The occurs whenever there is a nonrandom sample from a population and 2 measures that are imperfectly correlated. • Multiple-group threats to internal validity are criticisms that are likely to be raised when you have several groups in your study. Here, the key issue is the degree to which the groups are comparable before the study. Having multiple groups rules out all single-group threats to internal validity. The multiple-group case goes as follows: Group A # measure baseline, administer program, then measure outcomes (notated as R O X O); Group B (control group) # measure baseline, measure outcomes (notated as R O O). For these threats, the word “selection” is added in the front of each. To solve multiple group threats, randomly assign people to 2 groups or assign them non-randomly to be as equivalent as you can make them. o Selection threat: groups are not comparable at the beginning of the program; adding groups creates “noise” (example: Parents may “self-select” children into the program so that their children will get “ahead” – they will want to be in group A). This threat just means that there is something going on (an interaction) in the two groups that you didn’t know about. o Selection-history threat: any other event that occurs between pre- and post- test that the groups experience differently (example: Kids in group 1 pick up more math concepts because they watch Sesame Street). Here, the variable is present in both groups, but one group experiences the variable differently. o Selection-maturation threat: differential rates of normal development between pre- and post-test for the groups occur (example: Kids are learning at different rates, even without the program) o Selection-testing threat: differential effect on the posttest of taking the pretest (example: The test may have “primed” kids differently between the groups or they may have learned at different rates) o Selection-instrumentation threat: differential change in test used for each group from pre- and post-test; unlikely situation (example: Change due to different forms of the test being given differentially to groups; not due to the program) o Selection-mortality threat: differential nonrandom dropout between pre- and post- test (example: Kids drop out of study at different rates for each group) o Selection-regression threat: the group regresses towards the mean at a different rate because groups differ in extremity, meaning there are more outliers in one group. (example: Kids receiving program are disproportionately lower math scorers and therefore have greater regressions to the mean) • Social interaction threats to internal validity are threats that arise because social research is conducted in real-world human contexts; social pressures in research context can possibly lead to posttest differences not directly caused by the treatment itself. Social interaction threats are minimized by having multiple groups that are unaware of each other or by training administrators in the importance of keeping participants in the study and not trying to equalize the groups. o Diffusion or imitation of treatment: when the comparison group learns about the program and (possibly) make their own experience that imitates the program group’s experience; equalizes outcomes o Compensatory rivalry: when the comparison group develops a competitive attitude with the program group; equalizes outcomes o Resentful demoralization: when the comparison group knows what the other group is getting, and become discouraged/angry and drop out; or when the program group is assigned to an unpleasant condition then become discouraged and drop out; exaggerates outcomes o Compensatory equalization of treatment: when the administrators compensate one group for the perceived advantage of the other group How to come closer to establishing cause and effect in Public Health • Temporal precedence: when the cause happens before the effect. To establish causality, something has to happen temporarily. In a typical program evaluation, temporal precedence is taken care of because you intervene before you measure the outcome. • Covariation of the cause and the effect: both have to have a significant relationship; if cause, then effect, if not cause, then not effect. In a typical program evaluation, covariation of cause and effect is taken care of because you control the intervention, and you evaluate the impact of the program. The best way to do this is through a multiple-group program. rd • No plausible alternative explanations: this refers to the 3 variable problem. Internal validity is really the extent to which plausible alternative explanations have been ruled out. The best way to rule it out is statistically and to measure for things that might be related, or by adding a control group (a group comparable to your program group that differs only because it didn’t receive the program). In a typical program evaluation, this is the central issue of internal validity, but it is usually taken care of through your design. Session 24: The Nuts and Bolts of Successful Longitudinal Studies (CLS Guest Lecture) 12/9/13 6:30 PM The College Life Study (CLS) is a prospective, longitudinal study of young adults that began in 2004, and is funded by the NIH. The goal of the CLS is to gain a better understanding of the factors related to the initiation and discontinuation of both positive and negative health-related behaviors in young adults. Some topics covered include substance abuse, academic experience and achievements, demographics, and relationships. • Strategies to maintain high response rates include monetary incentives, consistent, frequent contact with respondents using a variety of contact methods, and a pleasant interview experience (involving flexibility in scheduling the interview in terms of timing, location, etc., and training the interviewers to be friendly, non- judgmental, grateful, able to tailor the tone of the interview to each respondent, and to stress confidentiality). • Research assistance comes from the interviewers (undergraduate and graduate students from all majors/programs; 7-10 employed at a time); they have to devote at least 15 hours a week, with 30- minute meetings with the recruitment coordinator once a week, and must commit to at least 2 semesters with the CLS. o Responsibilities of the research assistants include recruitment (contacting participants; every 3-4 days) using a variety of methods, scheduling and completing interviews, and help with office tasks. o Interviewers are trained using the following methods: ! Training materials are given to interviewers to review ! Two in-person group training sessions occur to go over recruitment methods, study protocol, interviewing skills, and a round-robin practice interview ! New interviewers observe at least 2 experienced staff members administer an interviewers, then are observed themselves while administering at least 2 interviews o Interviewers must be punctual, communicative, resourceful, respectful, honest, and flexible. • The importance of teamwork within the CLS impacts the whole process. Teamwork is required for covering interviews, making suggestions for recruiting respondents that are difficult to reach, discussing trickier parts of the interview, and venting after a difficult interview. • The CLS is confidential, not anonymous. Violations of confidentiality include disclosing a respondent’s identity, discussing a respondent’s responses to anyone outside of the research team, and leaving research materials in a place where someone else can read them. o Confidentiality is maintained through the staff’s completion of an online course that covers human subject research (CITI training), a federal certificate of confidentiality from the NIH (which is the highest validation of confidentiality), completed research materials being returned to the office within 2 days, storing information in secure locations, and giving participants 2 IDs – the CLS ID and a temporary ID. • CLS data goes from the interview/survey to field edit (which includes coding, looking for missing data; completed by the interviewer), to the quality assurance process (same process as field edit, done by a more experienced person), to scanning the paper interviews into a data reading program, to verifying/correcting the data entry (typing in handwritten answers), to data cleaning (checking for computer errors, adding additional data, checking “other” codes for common answers) Top resume and cover letter don’ts for life • Typos – always check, double check, and triple check for typos. • Formulaic, nonspecific cover letters (have specific reasons for each job) • “To whom it may concern” – only a “don’t” if a contact is provided • Incorrect information about the job or place of employment • Vague skills or very common skills (like, don’t put “Microsoft Office,” “responsible,” “hard worker,” “communication,” “research,” or “internet” as skills) • Fonts and formatting issues • Other recommendations o Don’t be afraid to email first with questions o Explain clearly and specifically why you’re interested in the position o Use you prior experience to your advantage, even if it doesn’t seem relevant o Know the name of your degree program! (in this case, it is a B.S. in Community Health) Session 25: Experimental Design – Introduction and Classification 12/9/13 6:30 PM Observational v. Experimental Studies • Observational studies are non-experimental; observations are made, but participants are not assigned to various conditions or groups. • Experimental studies assign individuals in the study to different groups to understand the impact of the group assignment on some outcome. The goal of an experimental study is to manipulate the independent variable while controlling extraneous variables. • Example: Does exposure to noise affect concentration? o Observational study: Observe the time it takes to do work on the loud floor of the library and a quiet floor of the library. o Experimental study: randomly assign people to noisy v. quiet settings, then measure the concentration rate of both groups. Random Selection v. Random Assignment: how do they differ? • Random selection is how you draw the sample of people for your study from a population. It impacts external validity, helping to insure that the sample is representative of the population. • Random assignment is how you assign the sample of people for your study from a population. This impacts internal validity by helping to insure that groups are comparable at the beginning of the study. Control Groups v. Comparison Groups • Control group: a group with no treatment • Comparison group: a group with some other treatment or a different level of treatment Placebo effects are used because participants have expectations, and those expectations are a potential confounding influence. Placebo effects mean that the mind and body are synergistic in terms of health. • Examples of placebos: sugar pills (used in clinical trials for a medication), and talking to a therapist when the therapist does not use the technique of interest (behavioral intervention) • To determine if a placebo effect exists, have 3 groups in the study: a treatment group, a no-treatment control group, and a placebo group. Experimental designs are the strongest regarding internal validity, but are difficult to carry out in the real world. The trade off to high internal validity can be reduced external validity (generalizability) due to “contrived” conditions of the experiment. Experimental design has 3 conditions: random assignment to groups, control groups, and manipulation of the independent variable (the variable that determines group assignment is the independent variable, in other words it is what you manipulate). • Note that the following terms are all interchangeable: treatment groups = intervention groups = experimental groups • Two-group experimental designs o The basic design is R X O, R O (2 groups, post-test only design), with no pretest required. It is most interested in determining if the groups are different after the program; compared by using a t-test or analysis of variance (ANOVA). This design is strong, but not the strongest, against all single group and multiple group threats except for the selection- mortality threat, and is susceptible to all social threats. o The two-group, pre-post test, randomized experiment (R O X O; R O O) has many advantages. You are able to quantify the change in the outcome variable and compare it between the 2 groups, able to confirm how well random assignment worked, and able to see whether or not some factor that existed prior to the experiment affected the outcome regardless of treatment. There is also the probability that variables that exist prior to the experiment interact with the treatment to produce different outcomes. ! Example: Hepatitis C treatment – why would you need a pre-test? " Did the IV drug users differ in their response to treatment compared to the individuals who got hepatitis C because of a blood transfusion? " Alcohol consumption during the course of the experiment could change the response to treatment because of its effects on immune function. Measure all things before. " How much did viral load decrease in the group that got the new drug compared to the group that received the “standard of care?” " Are you sure that the groups were “perfectly equivalent” in the level of severity of their disease before they started treatment? Probabilistic Equivalence doesn’t mean that the 2 groups will have identical means; it is achieved through random assignment. The Signal to Noise ratio: factorial designs enhance the signal while blocking and covariance designs reduce noise; used to classify experimental designs. The signal is the key variable of interest, and the noise is the random factors and variability in the series. The goal is for the signal to be high relative to the noise. Session 26: Experimental Design – Factorial, Randomized, Covariance, Switching Replications 12/9/13 6:30 PM Factorial Designs are signal-enhancing experimental designs, focusing on the setup of the program or treatment, that examine multiple variations of the treatment by crossing your variables in a chart. The advantages of factorial designs are that they give flexibility for exploring/enhancing treatment, they are efficient, and they are the only effective way to examine interaction effects. • A factor is a major independent variable in an experimental design. • A level is a subdivision of a factor (like an attribute to a variable) into components or features. • Example of factors and levels: Does time in instruction affect student achievement level? Does the setting in which instruction occurs make a difference? o Factor 1 is time in instruction. Level 1 is 1 hour/week, and level 2 is 4 hours/week. o Factor 2 is the setting of instruction. Level 1 is in pullout sessions (i.e., a discussion section) and level 2 is in class. • Main effect means that there is a relationship (effect = relationship) when consistent differences are seen between all levels of the factors. There are the same number of main effects as there are factors. If lines are flat (in graphs), there is no main effect. • Example of factorial design: Alcohol and perceived intoxication – the Ha is that alcohol increases perceived level of intoxication, and the Ho is that alcohol is not related to the perceived level of intoxication or alcohol decreases perceived intoxication. o Experimental design, 2 groups. Group 1 drinks one beer, group 2 drinks 5 beers. o Dependent variable is perceived intoxication rating; the independent variable is the level of alcohol consumption. o You are expecting a main effect of alcohol consumption on a perceived level of intoxication. o The result is that there is a main effect of alcohol, and that alcohol increases perceived intoxication. • Another example of a factorial design: Energy drinks and perceived intoxication. The Ha is that energy drinks in combination with alcohol decreases the perceived level of intoxication. The Ho is that energy drinks in combination with alcohol is not related to the perceived level of intoxication or that it increases the perceived intoxication o Experimental design, 2x2 (so, four groups). Group 1 has 1 beer and no energy drink, group 2 has one beer and one energy drink, group 3 has five beers and no energy drink, group 4 has five beers and one energy drink o Dependent variable is perceived intoxication rating, and the independent variable is the level of alcohol consumption and energy drink consumption (not level for ED, simply the presence of the ED). o Result: there was a main effect of the energy drinks on perceived intoxication. Energy drinks decreased the average rating of perceived intoxication. • An interaction effect is when differences on one factor depend on which level you are on in another factor (between factors, not levels), and is not limited to experimental designs. Statistical tables report all main effects and interactions. Not being able to talk about an effect on one factor without mentioning the other factor shows the presence of an interaction effect. Also, whenever lines are not parallel/when they cross, there is an interaction effect. o Example of an interaction effect (same example as above for main effects): Are the differences in rating of intoxication as a result of caffeine dependent upon the level of alcohol? Does the decrease in perceived intoxication only how up when people are really intoxicated? ! Result: energy drinks decrease the average rating of perceived intoxication only in the 5 beer condition. • A null outcome is a situation where the treatments have no effect (shown in graphs as when lines overlap each other). • Fully crossed factorial designs, where you look at all combinations of all factors, aren’t always necessary; incomplete factorial designs allow for a control/placebo group with no treatment. Covariance designs are noise-reducing experimental designs. • The analysis of covariance design (ANACOVA/ANCOVA) is a pre- /post- randomized experimental design. The preprogram measure is called a covariate because it is used with the outcome variable to remove variability/noise. Covariates are statistically controlled for in the data analysis to remove variability (the variables you are adjusting for in your study; attempting to remove the effects of one variable from another). The ANCOVA design “adjusts” post-test scores for variability on the covariate (pre-test). o Example of an ANCOVA design: Study on hypertension – group 1: R O X O, group 2: R O O. “X” is drug A, and the study is measuring stress in the pre-test and the post-test. The pre-test: group 1 had an average BP of 110 and stress level of 20, group 2 had an average BP of 110 and a stress level of 50. In the post-test: group 1’s average BP lowered to 70, group 2’s average BP stayed at 110. ! You can conclude from this experiment that drug A is successful. Blocking designs are noise/variance-reducing experimental designs. Randomized block designs are like stratified random sampling. They divide the sample into homogenous blocks, then implement the experiment within each block, or subgroup. They give more powerful estimates of treatment effects within blocks, and pool estimates across blocks to get a more efficient overall estimate. It is an analysis strategy, meaning it doesn’t affect participant interaction. • Needs a larger sample size to maintain statistical power (only possible drawback) Switching Replications Experimental Designs addresses the need to deny the program through random assignment of some participants to a no- program comparison group, and is used mainly in drug trials. In this design, implementation is repeated/replicated, meaning the 2 groups switch roles. This is used where programs are repeated at regular intervals. Switching replications experimental designs enhance organizational efficiency in resource allocation. • Example of a switching replications design: Treatment of hypertension, using drug A; group 1: R O X O, group 2: R O O X O o For group 1, the pre-existing blood pressure is 150, and after drug A it is 110. After removing drug A, the blood pressure goes back up to 150. In group 2, after not getting the drug, their blood pressure stays at 150. After they got drug A, their BP drops to 110, just like group A. The conclusion is that drug A works. o Alternative outcome: After drug A is removed, group 1’s blood pressure stays at 110. Drug A could either have a residual effect or be a fluke. But, it happened again in group 2. The best thing to do here would be to wait and measure again to see if the trend continued. o Alternative outcome: Group 1’s blood pressure stays the same after drug A is administered (creating a null effect). However, in group 2, it goes down. It could have been a fluke, or group 2 wasn’t matched on some type of variable. To move forward, figure out how the 2 groups are different. Session 27: Quasi-Experimental Design: Nonequivalent Groups Design, Regression- Discontinuity 12/9/13 6:30 PM Quasi-Experimental Designs are experimental designs that lack random assignment. Here, “quasi” means “seemingly,” or “apparently, but not really.” The research substitutes statistical “controls” for the absence of physical control of the experimental situation. • The nonequivalent-groups design is a pre-/post- test with a comparison group; it is a very common quasi-experimental design and is similar to the classic control design. However, participants cannot be randomly assigned to either the experimental or to the control group, or the researcher cannot control which group will get the treatment. Participants do not all have the same chance of being in the control or the experimental groups, or of receiving/not receiving the treatment. The notation for the basic design is N O X O; N O O (n = non-equivalent). • Example of a quasi-experimental design: A nutrition education program is being measured; some residents in Howard County were given the opportunity to participate in an after-school nutrition program. Measurements of dietary practices were available from a survey done prior to the program. After the program, the survey was administered again. o Controlling for threats to internal validity in quasi- experimental designs using this example: ! History: did some other current event affect the change in fast food consumption? Answer: No, because both groups experience the same current events. ! Maturation: were changes in fast food consumption due to normal developmental processes? Answer: No, because both groups experienced the same developmental problems. ! Regression: were subjects in group 1 just worse off in terms of fast food consumption to begin with? Answer: No. ! Selection: were the subjects self-selected into the experimental and control groups, which could affect the dependent variable? Answer: Yes, the groups in the program had parents who were much more concerned about their eating habits. So, we controlled for parental attitude statistically. ! Experimental mortality: did some subjects drop out? Did this affect the results? Answer: Yes, there were some students who did not attend all sessions because of sports, other family commitments, or because they didn’t like the program. We measured all of that and can look at the difference in outcome by level of attendance. ! Testing: did the pre-test scored affect the post-test? Answer: No, both groups were administered the same survey. ! Instrumentation: did measurement method change during research? Answer: No, both groups were administered the same survey. ! Design contamination: did the control group find out about the experimental treatment? Answer: Yes, some comparison families talked to the experimental families and got the materials from them. Regression-Discontinuity is a type of quasi-experimental design with the basic design as a pre-/post- test, 2 groups experiment. The unique method of assignment is that groups are based on their score on a pre-program measure (uses a cutoff). The advantage to RD designs is that it takes care of the ethical problem in assigning treatment to those who most need it, rather than randomly assigning to control/experience (experiment groups), and it is useful for determining a program’s effectiveness. The notation for a basic RD design is C O X O; C O O. • A well-implemented RD design is very strong in internal validity because it is almost the same as having randomized groups. It is very useful to explore casual hypotheses when randomization is not desirable or possible. • The disadvantages to RD designs are that they require large sample sizes, and are not very statistically powerful. • Example of a regression-discontinuity design: A new arthritis drug is developed, and the pre- and post- tests measure functional movement. The dependent variable is the functional movement scale (a lower score is worse). o Group 1 (treatment group) is given to people who score 0-50. Group 2 (control group) is given to people who score 51-100 on the scale. o The post-test shows a difference in regression line, producing discontinuity. More discontinuity means that the treatment is more effective. Session 28: Quasi-Experimental Design – Other Quasi-Experimental Designs 12/9/13 6:30 PM The interrupted time series design is best for studying policy or some type of event you’re measuring the impact of (example: Hurricane Katrina). There are several waves of observation before and after the introduction of the independent variable. Examples of interrupted time series designs include: • Implementation of a crackdown on speeding in a given state reduces the traffic fatality rate in that state. • Smoke-free laws (an analysis of Minnesota’s statewide smoke-free law) o A telephone survey was conduced to measure the dependent variable before the law went into effect and up to 18 months after the law went into effect (data collection) o Used 5 other states as comparisons to see if some other thing is going on that all are exposed to that you aren’t measuring for at the same time; varying the presence of the smoke-free law. o The main dependent variables were whether or not 1. people stopped smoking and 2. the perception of how many adults changed. o The study found no change in smoking behavior; they should have kept going with measurements or following up, or measured smoking initiation after the law was put in place. Other quasi-experimental designs are not as strong in internal validity, including: • The proxy-pretest design, which looks like a standard pre-post design with an important difference; the pre-test in this design is collected after the program is given. Here, a proxy variable is used to estimate where the groups would have been on the pre-test, with 2 variations: o The participants estimate their pre-test level, called the recollection proxy-pretest. This is not useful because people may not remember where they were or they may lie, but it is useful for recording perceptions of gain or change. o The use of archived records to stand in for the pretest (risk of instrumentation threat is present) • The double pre-test design (also known as a “dry run”), expressed as G1 N O O X O; G2 N O O O. This is important if you think that there is a maturation threat; it tries to minimize selection-maturity threats, and is strong in internal validity. • The separate pre-post samples design is used when the people you use for your pretest are different from the people you use for your posttest. (Example: Measuring customer satisfaction in 2 different agencies; you can’t measure the same people twice, but you’re still measuring customer satisfaction). This is useful when doing routine sample surveys in an organization or community. • The switching replications design is strong in terms of internal validity – it has 2 groups and 3 waves of measurement; both groups act as the control and the treatment group at some point in the study. Session 29: Designing Designs 12/9/13 6:30 PM Matching designs to research questions: is your research question descriptive? …Relational? …Causal? • Descriptive research questions are associated with observational studies, non-experimental designs, and might involve multiple measurements over time of the same variable to describe time trends. They can involve a description of the phenomenon or time trend among different groups. • Relational research questions are observational, non-experimental designs, are usually cross-sectional, and must involve the measurement of a “suspect” (an independent variable that is suspected to be related to the dependent variable). It can involve studying the relationship in the different groups. • Causal research questions can also be observational, but if they are, they have to be longitudinal. Quasi-experimental designs are OK, but it is best if it is an experimental design with random experiments. It should involve multiple time points since you can’t measure causality with only one point in time. It must involve the measurement of a suspect, done through experimental design, and can involve control or comparison groups. The inclusion of these groups can help you minimize the influence of other things (extraneous factors) that might influence the outcome of the study. Think of causality on a continuum; suspected # suggestive # very suggestive. o Suspected causality: something is going on between the 2 variables (longitudinal studies; non-experimental studies) o Suggested causality: experiments and quasi-experiments o Very suggestive causality: having successful experiments and experimental studies Minimizing threats to validity: • By argument before (a priori; best way/most convincing), or after (a posteriori): minimizing threats by arguing for your points; making up for mistakes. This is the most straightforward way to rule out threats. • By measurement and observation of possible or real threats and demonstrating that they are either minimal or nonexistent. • By design, including adding treatment or control groups, waves of measurement, etc. • By analysis, like 2-way factorial designs or ANCOVA • By preventative action, ruling out anticipated potential threats by taking some type of preventative action. Building a design involves the basic design elements, which are: • Time: casual relationships imply that there is a time lapse between the cause and the effect, indicated horizontally in experimental design notation. • Observation/measure: indicated with the symbol “O” in experimental design notation • Existence of a program, treatment, or intervention: in experimental designs, under control of researcher or naturally occurring; indicated by an “X” in design notation • Existence of a suspected risk factor in observational studies • Groups or individuals: in experimental designs, each group is indicated on a separate line • Method of assignment to groups Expanding a design: • Across time: including additional observation (pre-/post-tests) before or after the program; adding or removing programs • Across programs: like creating different levels • Across observations: multiple or redundant measures; measurements that theoretically should not be affected by the program; proxy measures (test scores for intelligence, etc.); recollected measures, or things measured administratively • Across groups: addition of a group to rule out specific threats to internal validity Part 5: Analysis and Write-Up (11/13-12/9)
More Less

Related notes for HLTH 200

Log In


Don't have an account?

Join OneClass

Access over 10 million pages of study
documents for 1.3 million courses.

Sign up

Join to view


By registering, I agree to the Terms and Privacy Policies
Already have an account?
Just a few more details

So we can recommend you notes for your school.

Reset Password

Please enter below the email address you registered with and we will send you a link to reset your password.

Add your courses

Get notes from the top students in your class.