When conducting multiple repertory grid interviews a common analysis strategy is to sort the constructs in semantically similar categories by means of inductive content analysis (Green, 2004; Jankowicz, 2004). Depending on the specific study, different goals may be relevant. The researcher may want to:
A common approach to achieve this is to conduct additional interviews until 1) and/or 2) are accomplished. However, when planning a repertory grid study, this is not satisfactory as the number of interviews to be conducted is not known a priori. As a remedy, a rule of thumb may be used. It has been documented that category saturation occurs between 15 and 25 interviews (Kerkhof, Cuppen, & Hisschemoller, 2009; Napier, Keil, & Tan, 2009; Tan & Hunter, 2002). However, this rule of thumb does not help to achieve goal 2).
Conducting simulations would be an appropriate way to assess the required sample size. However, for many applied researchers from the field of Personal Construct Psychology, where the repertory grid is frequently applied, this is not feasible. The
gridsampler GUI is designed to make such types of simulations an easy task. It allows to quickly conduct simulations necessary to get a rough estimate of the study specific required sample size without the need to program the simulation. Using the software the following generic type of questions can be answered:
“What is the probability to get a result where at least C percent of the categories contain a minimum of M attributes when using the sample size N?”
By reviewing the corresponding probabilities calculated by means of simulation (see below), an appropriate sample size can be determined.
To install the software from CRAN (recommended), use the following command:
If you prefer to install from GitHub, use the
To load and start the GUI type
The GUI displayed in the image below consists of a single display made up of three panels. In panel 1 and 2, the parameters for the specific study are set. In panel 3, the simulations are prompted and its results are displayed. In the following, the panels are explained.
The number of constructs elicited usually varies across persons (unless fixed by design). Depending on the field of study and context, a specific distribution of constructs can be assumed. For example, we may know from prior research, that for a study on car brands, individuals typically express between 4 and 8 constructs, with most individuals using 6 constructs. The particular expected distribution of elicited constructs across persons can be specified in panel 1. On the bottom of the panel, several presets for distribution are available. In most cases, a normal-like distribution will be suitable.
In this panel, the expected number of categories can be defined. The number of categories may vary across fields of research. In some fields, for example interpersonal relations, subjects tend to possess more constructs than for fields they have less experience with, for example when in comes to describing different types of snow. Each category is associated with a specific probability of being elicited. When the results for a single person are simulated, the attributes are drawn from this distribution without replacement, as the grid protocol assumes that identical constructs are not elicited twice in the same interview.
The rightmost panel is split into an upper and a lower part.
Upper part: In the upper part, a simulation can be prompted for a given sample size (number of subjects) by pressing “One Random Sample of Size N”. By pressing the button “R Random Samples of Size N” samples are repeatedly generated and the distribution of the results per category are indicated using selected percentiles. From the image, it can be inferred that the median number of occurrences of category 1 was 29, the 5th percentile at 23 and the 95th at 36. This gives the user a rough idea about the category counts to be expected.
Lower part: In the lower part of panel 3, this simulation is conducted for different samples sizes. The following parameters can be set:
The simulation is prompted by pressing “Simulate R Samples for Each N”. Right below the simulation parameters the drawing parameters can be defined. These are independent of the simulation itself and only determine which values are displayed. To redraw the graphic with different parameters press the “Redraw with New Settings” button.
All parameters can either be given as numeric values separated by commas or using the colon operator from R, i.e.
1, 2, 3. Additionally, it is also possible to use the
seq() function in the input field, e.g.
seq(20, 80, by = 5).
The following image shows an enlargement of the lower part of panel 3.