WEBVTT

1
00:00:01.159 --> 00:00:02.640
<v Speaker 1>All right, welcome back, folks.

2
00:00:03.399 --> 00:00:07.559
<v Speaker 2>So this is our last podcast on statistics, so we're

3
00:00:07.559 --> 00:00:10.560
<v Speaker 2>finally started to get past that point. We're heading into

4
00:00:10.599 --> 00:00:14.359
<v Speaker 2>ethics next. So let's start the two variables. Everything else

5
00:00:14.400 --> 00:00:18.000
<v Speaker 2>depends on independent variables. Is the manipulated factor with the

6
00:00:18.039 --> 00:00:21.440
<v Speaker 2>experiment or changes. Say a researcher wants to know if

7
00:00:21.480 --> 00:00:26.000
<v Speaker 2>EMDR reduces PTSD symptoms faster than prolonged exposure. The type

8
00:00:26.039 --> 00:00:29.239
<v Speaker 2>of therapy is the independent variable. It has to be

9
00:00:29.320 --> 00:00:33.960
<v Speaker 2>operationally defined, meaning it specified enough that another researcher could

10
00:00:33.960 --> 00:00:39.280
<v Speaker 2>replicated exactly, not just therapy eight sessions of manualized EMDR protocol,

11
00:00:39.759 --> 00:00:45.159
<v Speaker 2>standardized fidelity checklist, same therapist training level. The dependent variable,

12
00:00:45.200 --> 00:00:47.320
<v Speaker 2>the DV, is the measured outcome what changes as a

13
00:00:47.320 --> 00:00:50.520
<v Speaker 2>result of the independent variable and our PTSD study. That's

14
00:00:50.520 --> 00:00:53.359
<v Speaker 2>the change in score on a standardized instrument for PTSD.

15
00:00:54.359 --> 00:00:57.920
<v Speaker 2>The dependent variable needs to be reliable, valid, and sensitive

16
00:00:58.000 --> 00:01:01.600
<v Speaker 2>enough to detect real change. To use the blunt instrument

17
00:01:01.640 --> 00:01:04.359
<v Speaker 2>that can't move week a week, you'd miss a true

18
00:01:04.400 --> 00:01:09.079
<v Speaker 2>effect and think that therapy failed when it didn't. Operational

19
00:01:09.159 --> 00:01:12.280
<v Speaker 2>definitions apply to both sides. They define how constructs are

20
00:01:12.319 --> 00:01:16.640
<v Speaker 2>measured or manipulated in observable times, Depression becomes score on

21
00:01:16.680 --> 00:01:20.120
<v Speaker 2>the BDI two. Social support becomes number of perceived supportive

22
00:01:20.120 --> 00:01:23.920
<v Speaker 2>individuals rated seven or higher on a ten point scale.

23
00:01:24.400 --> 00:01:27.799
<v Speaker 2>This matter is because vague constructs can't be tested, replicated,

24
00:01:27.879 --> 00:01:32.799
<v Speaker 2>or falsified. Manipulation checks verify the independent variable actually had

25
00:01:32.799 --> 00:01:35.680
<v Speaker 2>its intended effect before you even look up the dependent variable.

26
00:01:36.280 --> 00:01:38.640
<v Speaker 2>If you're running a mood induction study and the sad

27
00:01:38.719 --> 00:01:43.239
<v Speaker 2>film clip didn't actually make people sadder, any downstream finding

28
00:01:43.280 --> 00:01:46.519
<v Speaker 2>about your mood and meaning and memory is meaningless, so

29
00:01:46.640 --> 00:01:49.120
<v Speaker 2>you insert a brief mood scale right after the induction

30
00:01:49.200 --> 00:01:52.680
<v Speaker 2>to confirm participants felt what you needed them to feel

31
00:01:52.719 --> 00:01:57.959
<v Speaker 2>before moving forward. Control conditions allow the comparison to baseline

32
00:01:58.040 --> 00:02:00.519
<v Speaker 2>or no treatment. This is what you let rule out

33
00:02:00.560 --> 00:02:04.439
<v Speaker 2>the possibility that people would have improved anyway. Some of

34
00:02:04.480 --> 00:02:08.560
<v Speaker 2>the types are include placebo, weight list, or treatment as usual.

35
00:02:10.199 --> 00:02:12.879
<v Speaker 2>A weight list control on a CBT for insomnia trial

36
00:02:12.919 --> 00:02:15.680
<v Speaker 2>holds people in their normal life circumstances while the treatment

37
00:02:15.719 --> 00:02:19.039
<v Speaker 2>group gets the intervention, so any difference between the groups

38
00:02:19.120 --> 00:02:23.479
<v Speaker 2>is more plausibly attributed to the therapy itself. Now controlling

39
00:02:23.520 --> 00:02:26.599
<v Speaker 2>extraaneous of variables, this is where four techniques do most

40
00:02:26.639 --> 00:02:31.319
<v Speaker 2>of the heavy lifting. Randomization assigns participants to groups by chance,

41
00:02:31.400 --> 00:02:35.759
<v Speaker 2>distributing uncontrolled variables evenly across conditions. It is the single

42
00:02:35.800 --> 00:02:40.159
<v Speaker 2>most powerful method for ensuring group equivalents because it doesn't

43
00:02:40.199 --> 00:02:42.599
<v Speaker 2>just balance the variables you thought of, it balances the

44
00:02:42.639 --> 00:02:47.159
<v Speaker 2>ones you didn't. Matching insures groups are equivalent on specific

45
00:02:47.240 --> 00:02:50.360
<v Speaker 2>variables like age or baseline severity. You'd use this when

46
00:02:50.360 --> 00:02:54.240
<v Speaker 2>your sample size as small or true randomization is infeasible,

47
00:02:54.319 --> 00:02:58.319
<v Speaker 2>say a specialty clinic with twelve treatment resistant patients. Matching

48
00:02:58.319 --> 00:03:01.439
<v Speaker 2>can be individual match pairs are frequency match groups, where

49
00:03:01.479 --> 00:03:04.879
<v Speaker 2>their overall distribution lines up even if individual pairs don't.

50
00:03:05.800 --> 00:03:09.319
<v Speaker 2>Blocking groups participants into blocks based on a known extraneous

51
00:03:09.400 --> 00:03:12.840
<v Speaker 2>variable like gender, and then randomizes within each block. If

52
00:03:12.840 --> 00:03:15.800
<v Speaker 2>you suspect gender might interact with your treatment effect, blocking

53
00:03:15.879 --> 00:03:18.879
<v Speaker 2>lets your control for that no known source of variation,

54
00:03:19.800 --> 00:03:25.520
<v Speaker 2>while still preserving randomization's benefits inside each block. Counterbalancing is

55
00:03:25.639 --> 00:03:29.840
<v Speaker 2>used within subject designs to control for order effects where

56
00:03:29.840 --> 00:03:35.759
<v Speaker 2>participants receive conditions in different sequences. Complete counterbalancing runs all

57
00:03:35.800 --> 00:03:39.759
<v Speaker 2>possible orders in Latin square is the efficient version, where

58
00:03:39.800 --> 00:03:42.800
<v Speaker 2>each condition appears once in each position across groups, so

59
00:03:42.840 --> 00:03:45.439
<v Speaker 2>you get most of the protection with far fewer participants.

60
00:03:46.400 --> 00:03:49.840
<v Speaker 2>Counterbalancing protects against practice effects where people get better just

61
00:03:49.879 --> 00:03:53.599
<v Speaker 2>from repetition, fatigue where people get worse from exhaustion, and

62
00:03:53.639 --> 00:03:55.879
<v Speaker 2>carry over, where the effects of one condition bleed into

63
00:03:55.879 --> 00:04:02.159
<v Speaker 2>the next. Blinding procedures address a different entirely expectancy, and

64
00:04:02.199 --> 00:04:05.639
<v Speaker 2>a single blind design participants are unaware of their group assignment.

65
00:04:06.159 --> 00:04:07.159
<v Speaker 1>This controls for.

66
00:04:07.280 --> 00:04:12.319
<v Speaker 2>Placebo effects, expectancy and demand characteristics where participants pick up

67
00:04:12.319 --> 00:04:15.159
<v Speaker 2>on what the study is looking for and unconsciously perform it,

68
00:04:16.040 --> 00:04:19.680
<v Speaker 2>and a double blind design both participants and experimenters are

69
00:04:19.720 --> 00:04:23.519
<v Speaker 2>blind to condition. This is critical in drug trial, psychotherapy,

70
00:04:23.519 --> 00:04:29.319
<v Speaker 2>comparison studies, and placebo controlled designs. It reduces experimental bias,

71
00:04:29.680 --> 00:04:33.439
<v Speaker 2>observer expectancy, and subtle queuing, where a researcher who knows

72
00:04:33.480 --> 00:04:36.000
<v Speaker 2>who got the real drug might unconsciously.

73
00:04:35.399 --> 00:04:36.920
<v Speaker 1>Rate their symptoms more favorably.

74
00:04:38.199 --> 00:04:41.319
<v Speaker 2>Let's look at the specialized designs, because the EPE trip

75
00:04:41.319 --> 00:04:43.439
<v Speaker 2>A P loves asking you to match the design to

76
00:04:43.519 --> 00:04:47.560
<v Speaker 2>threat the Solomon four group design. The Solomon four group

77
00:04:47.600 --> 00:04:51.920
<v Speaker 2>design combines pretest, post test and post test only designs

78
00:04:51.959 --> 00:04:55.920
<v Speaker 2>to control for pretest sensitization, also called the testing effect,

79
00:04:56.480 --> 00:04:59.879
<v Speaker 2>where simply taking a pretest changes how someone responds later.

80
00:05:00.560 --> 00:05:05.240
<v Speaker 2>It includes four groups pretest treatment post test, pretest, no

81
00:05:05.319 --> 00:05:09.279
<v Speaker 2>treatment post test, no pretest treatment post test, and no

82
00:05:09.399 --> 00:05:13.680
<v Speaker 2>pretest no treatment post test. This lets the research or

83
00:05:13.680 --> 00:05:16.720
<v Speaker 2>disentangle the effective treatment, the effect of testing itself, and

84
00:05:16.759 --> 00:05:20.240
<v Speaker 2>the interaction between the two. Imagine a study on self

85
00:05:20.240 --> 00:05:23.360
<v Speaker 2>compassion intervention with just filling out a self compassion scale

86
00:05:23.680 --> 00:05:27.519
<v Speaker 2>and four Ham primes people to think differently, and Solomon

87
00:05:27.600 --> 00:05:31.879
<v Speaker 2>design will catch that. The Latin square design controls for

88
00:05:31.959 --> 00:05:35.879
<v Speaker 2>other order effects. For order effects, the Latin square design

89
00:05:36.519 --> 00:05:39.360
<v Speaker 2>and repeated measure designs so where each treatment appears once

90
00:05:39.399 --> 00:05:41.920
<v Speaker 2>in each row and column, meaning each position, it's efficient

91
00:05:41.959 --> 00:05:46.720
<v Speaker 2>for controlling sequence effects with fewer participants than complete counterbalancing

92
00:05:46.720 --> 00:05:51.240
<v Speaker 2>would require three treatments, three orders, three groups, and you've

93
00:05:51.279 --> 00:05:57.319
<v Speaker 2>covered your sequence confound without needing six separate orderings. Fractional

94
00:05:57.439 --> 00:06:01.600
<v Speaker 2>factorial design get used when full factorial designs, which test

95
00:06:01.680 --> 00:06:05.879
<v Speaker 2>all independent variable combinations become too large to run. Only

96
00:06:06.000 --> 00:06:09.480
<v Speaker 2>they test only a subset of combinations while retaining interpretability.

97
00:06:09.959 --> 00:06:12.399
<v Speaker 2>This shows up in applied settings with many variables like

98
00:06:12.439 --> 00:06:16.319
<v Speaker 2>public health, messaging campaign, testing a message, tone, channel, and

99
00:06:16.360 --> 00:06:19.519
<v Speaker 2>framing all at once without needing. Every single combination represented

100
00:06:20.480 --> 00:06:23.519
<v Speaker 2>now single case experimental designs used in clinical and applied

101
00:06:23.519 --> 00:06:27.639
<v Speaker 2>psychology where a group designs simply aren't feasible. The ABA design,

102
00:06:27.759 --> 00:06:31.519
<v Speaker 2>also called reversal design, has three phases. As baseline B

103
00:06:31.639 --> 00:06:35.360
<v Speaker 2>is intervention and the second A is withdrawal. It measures

104
00:06:35.399 --> 00:06:39.519
<v Speaker 2>behavior change and reversibility whether the behavior returns toward baseline

105
00:06:39.519 --> 00:06:43.959
<v Speaker 2>once treatment is removed. The clear limitation is an ethical concern.

106
00:06:43.959 --> 00:06:46.279
<v Speaker 2>You're deliberately withdrawing a treatment that might be working.

107
00:06:47.160 --> 00:06:48.240
<v Speaker 1>Think of a token.

108
00:06:47.959 --> 00:06:50.639
<v Speaker 2>Economy for a child with disruptive behavior in a classroom.

109
00:06:50.680 --> 00:06:53.079
<v Speaker 2>You hesitate to pull an effective intervention just to prove

110
00:06:53.120 --> 00:06:59.079
<v Speaker 2>a point. Scientifically, the multiple baseline design introduces treatment at

111
00:06:59.079 --> 00:07:02.920
<v Speaker 2>different times across behaviors, settings, or individuals. It shows control

112
00:07:02.959 --> 00:07:06.879
<v Speaker 2>without needing withdrawal, which solves the ABA ethical problem. It's

113
00:07:06.920 --> 00:07:09.399
<v Speaker 2>appropriate for irreversible effects or when.

114
00:07:09.279 --> 00:07:10.480
<v Speaker 1>Removal would be harmful.

115
00:07:11.040 --> 00:07:13.720
<v Speaker 2>Think of a skill like reading fluency you can unlearn

116
00:07:13.800 --> 00:07:14.720
<v Speaker 2>to prove a point.

117
00:07:15.199 --> 00:07:16.959
<v Speaker 1>If a therapist introduces treatment for.

118
00:07:17.000 --> 00:07:19.600
<v Speaker 2>Behavior A at week two, behavior B at week four,

119
00:07:19.639 --> 00:07:22.639
<v Speaker 2>and behavior C at week six, and each behavior only

120
00:07:22.680 --> 00:07:24.560
<v Speaker 2>improves after its own treatment starts.

121
00:07:25.079 --> 00:07:28.199
<v Speaker 1>That staggered pattern lets you infer causality.

122
00:07:28.519 --> 00:07:33.040
<v Speaker 2>Single case designs emphasize repeated measurements, stability, and visual analysis,

123
00:07:34.279 --> 00:07:37.839
<v Speaker 2>meaning researchers look at graph data trends rather than relying

124
00:07:37.920 --> 00:07:41.319
<v Speaker 2>solely on inferential statistics. All of those connects back to

125
00:07:41.399 --> 00:07:44.800
<v Speaker 2>internal validity the extent to its changes in the dependent

126
00:07:44.879 --> 00:07:50.000
<v Speaker 2>variable are caused by the independent variable and not extraneous factors.

127
00:07:52.000 --> 00:07:54.759
<v Speaker 2>History refers to outside events that influence the outcome.

128
00:07:54.800 --> 00:07:56.319
<v Speaker 1>If a natural disaster hits.

129
00:07:56.120 --> 00:07:59.600
<v Speaker 2>The community in mid study, that history is contaminating the results.

130
00:08:01.000 --> 00:08:04.639
<v Speaker 2>Maturation is natural change over time. Kids get better at

131
00:08:04.680 --> 00:08:09.439
<v Speaker 2>emotional regulations simply by growing older. Testing is the effect

132
00:08:09.439 --> 00:08:13.720
<v Speaker 2>where the pretest itself affects the protest. This same testing

133
00:08:13.720 --> 00:08:18.519
<v Speaker 2>effect that Solomon design was built to catch. Instrumentation that

134
00:08:18.560 --> 00:08:21.120
<v Speaker 2>covers changes in the measurement tool or radar over the

135
00:08:21.160 --> 00:08:23.240
<v Speaker 2>course of the study. If a rater becomes more lenient

136
00:08:23.240 --> 00:08:26.439
<v Speaker 2>by the end of data collection, that instrumentation not real change.

137
00:08:27.360 --> 00:08:30.160
<v Speaker 2>Selection refers to group difference as a baseline before the

138
00:08:30.160 --> 00:08:32.559
<v Speaker 2>study even begins. If your treatment group happened to be

139
00:08:32.639 --> 00:08:36.240
<v Speaker 2>less severe going in, any improvement might just reflect who

140
00:08:36.279 --> 00:08:40.840
<v Speaker 2>ended up where attrition has dropped out that Alders group composition.

141
00:08:41.039 --> 00:08:43.559
<v Speaker 2>The most symptomatic participants drop out of the treatment are

142
00:08:43.600 --> 00:08:48.399
<v Speaker 2>in the remaining group, but looks artificially healthier. Regression to

143
00:08:48.440 --> 00:08:51.320
<v Speaker 2>the mean is when extreme scores move toward average over time,

144
00:08:51.399 --> 00:08:55.039
<v Speaker 2>regardless of treatment. Someone who scored unusually high on anxiety

145
00:08:55.080 --> 00:08:57.840
<v Speaker 2>measure at intake is statistically likely to score lower next time,

146
00:08:57.879 --> 00:09:03.279
<v Speaker 2>even with no intervention at all. Good design eliminates or

147
00:09:03.320 --> 00:09:07.039
<v Speaker 2>controls for these threats using random assignments, control groups, blinding

148
00:09:07.120 --> 00:09:11.519
<v Speaker 2>manipulation checks, and consistent measurement procedures, meaning the same protocol,

149
00:09:12.080 --> 00:09:15.519
<v Speaker 2>same instruments, same conditions, apply the same way across the

150
00:09:15.679 --> 00:09:16.360
<v Speaker 2>entire study.

151
00:09:17.559 --> 00:09:19.320
<v Speaker 1>Well that's the episode. Know of those

152
00:09:19.399 --> 00:09:22.039
<v Speaker 2>Terms, but more importantly, know what each problem is solving.